CN113641802B

CN113641802B - Interaction method, model training method, device, equipment and storage medium

Info

Publication number: CN113641802B
Application number: CN202010394673.5A
Authority: CN
Inventors: 孙一峤; 田俊峰; 许欢; 肖文明
Original assignee: Alibaba Group Holding Ltd
Current assignee: Alibaba Group Holding Ltd
Priority date: 2020-05-11
Filing date: 2020-05-11
Publication date: 2025-03-14
Anticipated expiration: 2040-05-11
Also published as: CN113641802A

Abstract

The embodiment of the invention provides an interaction method, a model training method, a device, equipment and a storage medium, wherein the method comprises the following steps: firstly, acquiring an interaction record aiming at a target object and identification information of an interaction main body, wherein the interaction record is used for describing interaction actions generated by the interaction main body on the target object. And then, determining the expected interaction action according to the identification information of the interaction main body and the interaction record, and finally generating a response record describing the expected interaction action to realize interaction. Therefore, the automatic interaction method can improve the transaction efficiency. Meanwhile, when the response record is generated, the current interaction action of the interaction main body and the identification information for reflecting the interaction habit of the interaction main body are considered, so that the generated response record is more in line with the interaction habit of the interaction main body, and the success rate of the transaction is improved.

Description

Interaction method, model training method, device, equipment and storage medium

Technical Field

The present invention relates to the field of computer technologies, and in particular, to an interaction method, a model training method, a device, equipment, and a storage medium.

Background

For online transactions of users, they may be considered to be an interactive behavior from person to person. In the interaction process, the interaction parties, namely the buying and selling parties, can communicate with each other according to the details, prices and other contents of the commodity. Generally, the buyer and the seller need to save the online state at the same time and communicate for a plurality of rounds, so that the transaction is possible.

However, in the process of transaction, sellers may need to deal with a plurality of buyers, and the communication pressure of sellers is high, so that the transaction efficiency and the transaction success rate are not high. Therefore, how to ensure the success rate of the transaction is a urgent problem to be solved.

Disclosure of Invention

In view of this, the embodiments of the present invention provide an interaction method, a model training device, a device and a storage medium, which are used to ensure the success rate of transactions.

In a first aspect, an embodiment of the present invention provides an interaction method, including:

Acquiring identification information of an interaction subject and an interaction record aiming at a target object, wherein the interaction record is used for describing interaction actions;

determining an expected interaction action according to the identification information and the interaction record;

And generating a response record corresponding to the interaction record according to the expected interaction action.

In a second aspect, an embodiment of the present invention provides an interaction device, including:

The system comprises an acquisition module, a target object acquisition module and a display module, wherein the acquisition module is used for acquiring identification information of an interaction main body and an interaction record aiming at the target object, and the interaction record is used for describing interaction actions;

The action determining module is used for determining expected interaction actions according to the identification information and the interaction records;

And the generating module is used for generating a response record corresponding to the interaction record according to the expected interaction action.

In a third aspect, an embodiment of the present invention provides an electronic device, including a processor and a memory, where the memory is configured to store one or more computer instructions, and the one or more computer instructions implement the interaction method in the first aspect when executed by the processor. The electronic device may also include a communication interface for communicating with other devices or communication networks.

In a fourth aspect, embodiments of the present invention provide a non-transitory machine-readable storage medium having stored thereon executable code, which when executed by a processor of an electronic device, causes the processor to at least implement the interaction method according to the first aspect.

In a fifth aspect, an embodiment of the present invention provides an interaction method, including:

displaying an interactive interface comprising interactive buttons;

responding to click operation triggered by an interaction main body on the interaction button, and acquiring an interaction record aiming at a target object;

Determining an expected interaction action according to the interaction record and the identification information of the interaction main body;

and displaying a response record for describing the expected interaction in the interaction interface.

In a sixth aspect, an embodiment of the present invention provides an interaction device, including:

the display module is used for displaying an interactive interface comprising interactive buttons;

the acquisition module is used for responding to click operation triggered by the interaction main body on the interaction button and acquiring interaction records aiming at the target object;

The action determining module is used for determining expected interaction actions according to the interaction records and the identification information of the interaction main body;

The display module is further used for displaying a response record for describing the expected interaction action in the interaction interface.

In a seventh aspect, an embodiment of the present invention provides an electronic device, including a processor and a memory, where the memory is configured to store one or more computer instructions, and the one or more computer instructions implement the interaction method in the fifth aspect when executed by the processor. The electronic device may also include a communication interface for communicating with other devices or communication networks.

In an eighth aspect, embodiments of the present invention provide a non-transitory machine-readable storage medium having stored thereon executable code, which when executed by a processor of an electronic device, causes the processor to at least implement the interaction method according to the fifth aspect.

In a ninth aspect, an embodiment of the present invention provides an interaction method, including:

displaying an interactive interface comprising interactive buttons;

responding to clicking operation triggered by a first interaction party on the interaction button, and acquiring respective identification information of the first interaction party and the second interaction party and a first interaction record aiming at a target object;

Determining an expected interaction action of the first interaction party according to the first interaction record and the identification information of the second interaction party;

displaying a first response record describing an intended interaction of the first interactor;

responding to click operation triggered by the second interaction party on an interaction button, and determining expected interaction action of the second interaction party according to the first interaction record, the first response record and the identification information of the first interaction party;

A second answer record describing the intended interaction of the second interaction partner is displayed.

In a tenth aspect, an embodiment of the present invention provides an interaction device, including:

the acquisition module is used for responding to clicking operation triggered by a first interaction party on the interaction button and acquiring respective identification information of the first interaction party and the second interaction party and a first interaction record aiming at a target object;

The first action determining module is used for determining the expected interaction action of the first interaction party according to the first interaction record and the identification information of the second interaction party;

The display module is further used for displaying a first response record for describing expected interaction actions of the first interaction party;

The second action determining module is used for responding to the click operation triggered by the second interaction party on the interaction button and determining the expected interaction action of the second interaction party according to the first interaction record, the first response record and the identification information of the first interaction party;

the display module is further configured to display a second answer record for describing an expected interaction of the second interaction party.

In an eleventh aspect, an embodiment of the present invention provides an electronic device, including a processor and a memory, where the memory is configured to store one or more computer instructions, and the one or more computer instructions implement the interaction method in the ninth aspect when executed by the processor. The electronic device may also include a communication interface for communicating with other devices or communication networks.

In a twelfth aspect, embodiments of the present invention provide a non-transitory machine-readable storage medium having executable code stored thereon, which when executed by a processor of an electronic device, causes the processor to at least implement the interaction method according to the ninth aspect.

In a thirteenth aspect, an embodiment of the present invention provides a model training method, including:

receiving an interaction record of a first interaction party corresponding to a history object, the interaction record comprising a plurality of records generated by the first interaction party and a second interaction party, the interaction record being input by the second interaction party;

determining interaction characteristics of the first interaction party according to the interaction records;

Determining action prediction parameters according to the interaction characteristics;

and outputting the action prediction parameters so that the second interaction party obtains prediction model parameters for the first interaction party according to the action prediction parameters.

In a fourteenth aspect, an embodiment of the present invention provides a model training apparatus, including:

the system comprises a receiving module, a processing module and a processing module, wherein the receiving module is used for receiving an interaction record of a first interaction party corresponding to a history object, the interaction record comprises a plurality of records generated by the first interaction party and a second interaction party, and the interaction record is input by the second interaction party;

The characteristic determining module is used for determining the interaction characteristic of the first interaction party according to the interaction record;

The parameter determining module is used for determining action prediction parameters according to the interaction characteristics;

and the output module is used for outputting the action prediction parameters so that the second interaction party obtains the prediction model parameters aiming at the first interaction party according to the action prediction parameters.

In a fifteenth aspect, an embodiment of the present invention provides an electronic device, including a processor and a memory, where the memory is configured to store one or more computer instructions, where the one or more computer instructions, when executed by the processor, implement the model training method in the thirteenth aspect. The electronic device may also include a communication interface for communicating with other devices or communication networks.

In a sixteenth aspect, embodiments of the present invention provide a non-transitory machine-readable storage medium having executable code stored thereon, which when executed by a processor of an electronic device, causes the processor to at least implement the model training method according to the thirteenth aspect.

In a seventeenth aspect, an embodiment of the present invention provides a model training method, including:

Acquiring an interaction record corresponding to a history object, wherein the interaction record comprises a plurality of records generated by a first interaction party and a second interaction party;

And determining an action prediction model of the second interaction party aiming at the first interaction party according to the interaction characteristics.

In an eighteenth aspect, an embodiment of the present invention provides a model training apparatus, including:

the system comprises an acquisition module, a storage module and a storage module, wherein the acquisition module is used for acquiring an interaction record corresponding to a history object, and the interaction record comprises a plurality of records generated by a first interaction party and a second interaction party;

And the model determining module is used for determining an action prediction model of the second interaction party for the first interaction party according to the interaction characteristics.

In a nineteenth aspect, an embodiment of the present invention provides an electronic device, including a processor and a memory, where the memory is configured to store one or more computer instructions, and the one or more computer instructions, when executed by the processor, implement the model training method in the seventeenth aspect. The electronic device may also include a communication interface for communicating with other devices or communication networks.

In a twentieth aspect, embodiments of the present invention provide a non-transitory machine-readable storage medium having executable code stored thereon, which when executed by a processor of an electronic device, causes the processor to at least implement the model training method as described in the seventeenth aspect.

According to the interaction method provided by the embodiment of the invention, the interaction record aiming at the target object and the identification information of the interaction main body are acquired, wherein the interaction record is used for describing the interaction action of the interaction main body on the target object. And then, according to the identification information of the interaction main body and the interaction record, determining the expected interaction action, and finally generating a record describing the expected interaction action, namely, a response record corresponding to the interaction record, so as to realize interaction.

According to the above description, in one aspect, the present invention provides an automatic interaction manner, that is, automatically generating an interaction action of the next step of the interaction subject, that is, an expected interaction action, according to the actions described in the interaction record. On the other hand, when the response record is generated, the current interaction action of the interaction main body and the identification information of the interaction main body are considered. The identification information can embody the interaction habit of the interaction main body, so that the generated response record is more accordant with the interaction habit of the interaction party and has more pertinence.

When the interaction is specifically commodity transaction in the background technology, the automatic transaction method can improve the transaction efficiency, and meanwhile, the generated response record accords with the transaction habit of the first interaction party, so that the transaction success rate can be greatly improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flow chart of an interaction method provided by an embodiment of the present invention;

FIG. 2 is a schematic diagram of an interaction record according to an embodiment of the present invention;

FIG. 3a is a flowchart of an alternative motion prediction method according to an embodiment of the present invention;

FIG. 3b is a flow chart of another alternative motion prediction method according to an embodiment of the present invention;

FIG. 4 is a flowchart of another interaction method according to an embodiment of the present invention;

FIG. 5 is a flowchart of another interaction method according to an embodiment of the present invention;

Fig. 6 is a schematic diagram of an interaction method according to an embodiment of the present invention applied in a transaction scenario;

FIG. 7 is a flowchart of a model training method according to an embodiment of the present invention;

FIG. 8 is a flowchart of another model training method according to an embodiment of the present invention;

FIG. 9 is a flowchart of a method for determining a motion prediction model according to an embodiment of the present invention;

FIG. 10 is a flowchart of another method for determining a motion prediction model according to an embodiment of the present invention;

FIG. 11 is a flowchart illustrating a further method for determining a motion prediction model according to an embodiment of the present invention;

Fig. 12 is a schematic structural diagram of an interaction device according to an embodiment of the present invention;

Fig. 13 is a schematic structural diagram of an electronic device corresponding to the interaction device provided in the embodiment shown in fig. 12;

FIG. 14 is a schematic structural diagram of another interactive device according to an embodiment of the present invention;

fig. 15 is a schematic structural diagram of an electronic device corresponding to the interaction device provided in the embodiment shown in fig. 14;

FIG. 16 is a schematic structural diagram of yet another interactive device according to an embodiment of the present invention;

FIG. 17 is a schematic structural diagram of an electronic device corresponding to the interactive device provided in the embodiment shown in FIG. 16;

FIG. 18 is a schematic structural diagram of a model training device according to an embodiment of the present invention;

FIG. 19 is a schematic structural diagram of an electronic device corresponding to the model training apparatus provided in the embodiment shown in FIG. 18;

FIG. 20 is a schematic structural diagram of another model training apparatus according to an embodiment of the present invention;

Fig. 21 is a schematic structural diagram of an electronic device corresponding to the model training apparatus provided in the embodiment shown in fig. 20.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The terminology used in the embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in this application and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise, the "plurality" generally includes at least two, but does not exclude the case of at least one.

It should be understood that the term "and/or" as used herein is merely an association relationship describing the associated object, and means that there may be three relationships, e.g., a and/or B, and that there may be three cases where a exists alone, while a and B exist together, and B exists alone. In addition, the character "/" herein generally indicates that the front and rear associated objects are an "or" relationship.

The words "if", as used herein, may be interpreted as "at" or "when" depending on the context, or "in response to a determination" or "in response to an identification". Similarly, the phrase "if determined" or "if identified (stated condition or event)" may be interpreted as "when determined" or "in response to a determination" or "when identified (stated condition or event)" or "in response to an identification (stated condition or event), depending on the context.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a product or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such product or system. Without further limitation, an element defined by the phrase "comprising one of the elements" does not exclude the presence of additional identical elements in a commodity or system comprising the element.

Before the detailed description of the interaction method provided by the embodiment of the invention, the actual meaning of the interaction method may be described in an exemplary manner:

in daily life, various interaction scenes can exist, and as mentioned in the background art, one common scene can be that both interaction parties respectively use terminal equipment to conduct online transactions. And the transaction can be either a secondhand transaction of the commodity or a non-secondhand transaction. At this time, the interaction parties can directly communicate the contents such as detailed information and price of the transaction object through the application program installed on the terminal equipment. However, the original interaction mode can be completed only by the simultaneous online of the two interaction parties, and if one party is offline or one party cannot reply in time, the transaction efficiency and success rate are greatly reduced. At this time, the interaction method provided by the invention can be used to avoid the problems.

In another interaction scenario, user A may interact with the intelligent robot. But the intelligent robot outputs an answer sentence without emotion color. In order to improve the interactive experience of the user A, the interactive method provided by the invention can be used, and at the moment, the language and expression mode of the answer sentence output by the intelligent robot can be the same as those of the other user B familiar to the user, so that the user A has a feeling of talking with the user B.

Of course, the present invention is not limited to use scenarios, and any scenario requiring automatic interaction may be provided by the present invention, except for the above scenario.

Some embodiments of the present invention are described in detail below with reference to the accompanying drawings. In the case where there is no conflict between the embodiments, the following embodiments and features in the embodiments may be combined with each other. In addition, the sequence of steps in the method embodiments described below is only an example and is not strictly limited.

The following embodiments are described taking the transaction scenario mentioned above as an example. In this scenario, the interaction is a transaction action generated by the interaction body on the target object, which may be a commodity. The transaction action includes any one of a cut-and-price action on the target object, a query action on attribute information of the target object, a transaction success action on the target object, and a transaction failure action on the target object.

Based on the foregoing, fig. 1 is a flowchart of an interaction method according to an embodiment of the present invention, where the interaction method according to the embodiment of the present invention may be executed by an interaction device. It is understood that the interactive device may be implemented in software, or a combination of software and hardware. The interaction device in this embodiment and the embodiments described below may specifically be a terminal device used by an interaction entity, such as a mobile phone or the like. As shown in fig. 1, the method comprises the steps of:

s101, acquiring identification information of an interaction subject and an interaction record aiming at a target object, wherein the interaction record is used for describing interaction actions.

The interaction device used by the interaction body can be provided with a transaction Application (APP). The interaction body can initiate transaction on any commodity to be transacted, namely a target object, on the transaction APP. After the transaction is initiated, the interaction body can start a dialogue for the target object, and the dialogue content can be synchronously displayed on the interaction device used by the interaction body.

The dialogue content may include an interaction record generated by the interaction body and aimed at the target object. Alternatively, the interaction records may be presented in a plurality of types of text, video, pictures. In terms of content, the interaction record may be detailed information of the query target object, such as a model number, a size, etc., or a discussion price, etc.

In the process of the interaction body performing the dialogue, assuming that the interaction body initiates the transaction for the target object at the time T1, the dialogue content for the target object generated by the interaction body is the interaction record in the time period from the time T1 to the current time T2. The interaction device used by the interaction agent may obtain and display this interaction record. For example, at time T2, the interaction record generated by the interaction agent for the target object may be shown in fig. 2.

Meanwhile, after the interaction main body initiates the transaction, the interaction equipment used by the interaction main body can also obtain the user identification of the interaction main body, and further inquire the identification information of the interaction main body according to the user identification. The user identification may be specifically represented as a user ID, and the identification information is used to reflect a transaction habit of the first interaction party.

S102, determining expected interaction according to the identification information and the interaction record.

For the interaction record obtained by the interaction device, which has been described as the interaction action generated by the interaction body in the period from the time T1 to the time T2, the interaction record shown in fig. 2 is continued to be accepted, for the interaction action, the following can be understood:

During this time period, the interaction by the interacting body may be a cut-off price-disagree with a cut-off price. At this time, the interactive device determines the expected interactive action according to the interactive action and the identification information of the interactive body. This intended interaction may be considered to be the interaction that, in the current transaction state, conforms to the transaction habits of the interaction partner, is most accepted by the interaction partner, and is also the most likely interaction for successful transactions.

S103, generating a response record corresponding to the interaction record according to the expected interaction action.

After the interactive device obtains the expected interaction, the record describing the expected interaction is further generated, that is, the response record corresponding to the interaction record. In practical applications, the interactive records and the response records are usually expressed in the form of text sentences, and for the generation of the response records, alternatively, the interactive device already contains sentence templates corresponding to different interactive actions, and the generation of the response records can be completed through inquiry. The generated response record is simultaneously displayed in the interactive equipment used by the interactive main body, namely, the automatic response to the interactive record is completed.

In this embodiment, an interaction record for a target object and identification information of an interaction subject are acquired, where the interaction record is used to describe an interaction action generated by the interaction subject on the target object. And then, determining the expected interaction action according to the identification information of the interaction main body and the interaction record, and finally generating a response record describing the expected interaction action to realize interaction. Therefore, the automatic interaction method can improve the transaction efficiency. Meanwhile, when the response record is generated, the current interaction action of the interaction main body and the identification information for reflecting the interaction habit of the interaction main body are considered, so that the generated response record is more in line with the interaction habit of the interaction main body, and the success rate of the transaction is improved.

In practical applications, the interaction body mentioned in the foregoing embodiment may specifically include a first interaction party and a second interaction party. At this point, the determination of the expected interaction is actually that the expected action of the second interaction party is determined according to the identification information of the first interaction party and the interaction record describing the current interaction action of the first interaction party. The expected interaction is the interaction which accords with the transaction habit of the first interaction party, is most accepted by the first interaction party and is most likely to lead the transaction to be successful in the current transaction state. In the embodiments described below, the first party may be considered a buyer and the second party a seller. And the interactive devices mentioned in the following embodiments are all interactive devices used by the second interactive party unless otherwise specified.

Based on the above description, optionally, the determination of the expected interaction may be achieved by means of a predictive model. Specifically, the identification information of the interaction main body, that is, the identification information of the first interaction party is obtained, and at this time, the interaction device may determine, according to the identification information, a motion prediction model of the second interaction party with respect to the first interaction party. The identification information may be considered as a model parameter of the motion prediction model, and the motion prediction model corresponding to the first interaction party may be generated directly according to the identification information of the first interaction party. And then inputting interaction records generated by the interaction parties aiming at the target object in the time period from the time T1 to the time T2 into the action prediction model so as to output the expected interaction action of the second interaction party, namely the expected transaction action, through the action prediction model.

In practical applications, the motion prediction model may be various, and the process of predicting the expected interaction motion of the second interaction party by using the motion prediction model is different. An alternative way of motion prediction is shown in fig. 3 a:

s201, inputting the interaction records, the attribute information of the target object and the transaction characteristics of the first interaction party and the second interaction party into an action prediction model, and extracting the characteristics of the interaction records, the attribute information and the transaction characteristics through a characteristic extraction layer of the action prediction model.

S202, inputting the extracted characteristics into an output layer of the action prediction model to output the expected transaction action of the second interaction party through the output layer.

And inputting the interaction record, the attribute information of the target object and the transaction characteristics of the first interaction party and the second interaction party into the action prediction model together so as to output the expected transaction action of the second interaction party on the target object through the action prediction model. The motion prediction may be a convolutional neural network (Recurrent Neural Network, RNN for short) model. The training process for the motion prediction model may be described in detail with reference to the embodiment shown in fig. 11 below.

Specifically, the feature extraction layer in the action prediction model may perform feature extraction on the transaction features of the interaction record, the attribute information of the target object, and the interaction parties. The interaction record describes the current interaction state of the target object and the current transaction action of the first interaction party. And then inputting the extracted characteristics into an output layer of the action prediction model so as to output the expected transaction action through the output layer.

Wherein the current interaction state of the target object may be an action sequence consisting of transaction actions described by each of the interaction records. Continuing with the example shown in FIG. 2, at the current time T2, the current interaction state depicted in the interaction record may be cut-off (first interactors) -disagree with cut-off (second interactors) -cut-off (first interactors). The current transaction action of the first interactive party is a cut price and the current transaction action of the second interactive party is a disagreement with the cut price.

The attribute information of the target object may include detail information of the target transaction object and/or transaction competitiveness of the target transaction object. The detail information may be, for example, the size, model, type, etc. of the commodity. It is easy to understand that, on the transaction platform, there may be multiple sellers for the same target transaction object, and the second interaction party is one of the multiple sellers, and the transaction competitiveness may be expressed as that the time T2 is on the transaction platform, and the multiple sellers have intermediate prices for the target transaction object/prices for the target transaction object by the second interaction party. The higher the ratio, the higher the transaction competitiveness.

It should be noted that, after the first interaction party initiates the transaction on the target transaction object, the interaction device may begin to collect the data, so as to obtain attribute information of the target transaction object.

For intermediate prices, the interactive device may alternatively determine how many transaction objects are the same as the target transaction object on the transaction platform by identifying a picture of each transaction object on the transaction APP. If the transaction object is purchased on the transaction APP, the second interaction party can directly use the secondary transaction function provided on the APP to re-line the transaction object, and the interaction device can directly obtain the detailed information of the transaction object, so as to determine whether the detailed information is the same as the target transaction object. After determining the transaction object which is the same as the target transaction object on the transaction platform, the intermediate price of the target transaction object can be further obtained.

The interaction characteristics of the first interaction party may include at least one of bargained capability, transaction frequency, transaction efficiency, and transaction success rate of the first interaction party within a preset period of time, such as a year. The same is true of the interaction characteristics of the second interaction party, and will not be described in detail herein.

The transaction efficiency may be considered as transaction endurance, i.e. the average number of interaction rounds of the first interaction party over a preset period of time. A question and answer between a first party and a second party may be considered a round of interaction.

Bargained capability, i.e. average bargain price/average bid price of the transaction object for which the first interaction party is successful in the transaction within a preset period of time. The greater the ratio, the higher the transaction capability when the first interactive party is a buyer and the smaller the ratio, the higher the transaction capability when the first interactive party is a seller.

The transaction frequency may be specifically considered as transaction liveness, that is, the transaction amount that the first interaction party initiates a transaction for how many transaction objects on the transaction platform within a preset period of time, that is, the transaction total amount, where the transaction success is included and the transaction failure is included.

The transaction success rate may be specifically regarded as the transaction proficiency, that is, the amount of work done by the first interactive party per total amount of transaction in a preset period of time.

It should be noted that the transaction characteristics described above may be updated regularly. And after the first interaction party initiates the transaction to the target transaction object, the interaction equipment can directly acquire the transaction.

In this embodiment, in the process of predicting the expected transaction action of the second interaction party, attribute information of the target object and transaction characteristics capable of reflecting the transaction habit of the first interaction party are used, so that the predicted expected transaction action can more conform to the transaction habit of the first interaction party, and the transaction success rate is improved.

Alternatively, the interaction record and the attribute information of the target object are input into the action prediction model to predict the expected transaction action of the second interaction party through the action prediction model. Alternatively, the motion prediction model may be a Markov decision (Markov Decision Process, MDP) model. The model parameters of the motion prediction model may be determined according to the transaction characteristics of the first interaction party and the attribute information of the historical transaction objects of the first interaction party, and the determination process of the parameters may be described with reference to the embodiment shown in fig. 10.

Wherein the attribute information of the historical transaction object comprises the detail information of the historical transaction object and/or the transaction competitiveness of the historical transaction object. The historical transaction objects are all transaction objects of which the first interaction party succeeds or fails in transaction within a preset time period. The specific content of the attribute information and transaction characteristics may be found in the description of the embodiment shown in fig. 3 a.

In this embodiment, similar to the embodiment shown in fig. 3a, attribute information of the target object and transaction characteristics capable of reflecting the transaction habit of the first interaction party are used in the action prediction process, so that the predicted expected transaction action more accords with the transaction habit of the first interaction party, and the transaction success rate is improved.

In addition to the two ways described above, yet another alternative is to act as a prediction, as shown in fig. 3 b:

S203, the current transaction action of the first interaction party described in the interaction record is obtained.

S204, determining the expected transaction action of the second interaction party according to the current transaction action and a prediction equation in the action prediction model, wherein the prediction equation is determined according to the transaction characteristic of the first interaction party.

After the interaction record is acquired, the interaction device acquires the current transaction action of the first interaction party because the current interaction action of the first interaction party, namely the current transaction action, is described in the interaction record. Then, optionally, an alternative transaction action of the second interaction party may be determined from the current transaction action and a predictive equation in the action predictive model. The motion prediction model may be a negotiation decision (Negotiation Decision Function, abbreviated as NDF) model.

If the preset condition is met between the current transaction action and the alternative transaction action, and the fact that the bid of the current first interaction party is in contact with the bid of the first interaction party is indicated, the satisfaction degree of the first interaction party is lower, and the interaction equipment determines the alternative transaction action as the expected interaction action of the second interaction party. Wherein the alternative transaction action is generally any other transaction action that is either a non-transaction success or failure.

If the preset condition is not met between the current transaction action and the alternative transaction action, and the fact that the bid of the current first interaction party is used for making an exchange is indicated, the satisfaction degree of the first interaction party is higher, the interaction equipment can directly generate a response record used for indicating that the transaction is successful, and the target object is used for making an exchange with the bid of the current first interaction party.

For the determination of whether the current transaction action and the alternative transaction action meet the preset conditions, optionally,

The respective utility values of the current transaction action and the alternative transaction action are respectively determined according to utility equations. The utility value of the current transaction action reflects the satisfaction degree of the first interaction party after the target object is transacted with the current transaction action. For example, when the current transaction action of the first interactive party is a cut price, the smaller the number of interactive rounds to which the cut price action belongs, the more the current price differs from the bid price of the second interactive party, the higher the utility value corresponding to the cut price action.

And determining whether a preset condition is met between the current transaction action and the alternative transaction action according to the utility value and the utility loss parameter, wherein the utility loss parameter can be determined according to the transaction characteristic of the first interaction party. Wherein, the current transaction action may be denoted as σ1, the utility value of the current transaction action may be denoted as U (σ1), the alternative transaction action may be denoted as σ2, the utility value of the alternative transaction action may be denoted as U (σ2), and the utility loss parameter may be denoted as δ. If U (sigma 1) is less than or equal to delta U (sigma 2), the preset condition is met between the two actions, otherwise, the preset condition is not met.

In the transaction process, the cut price is the most common transaction action, and on one hand, for the utility equation used, the simplest case can be considered as the difference between the historical average transaction price corresponding to the historical transaction object of the first interaction party and the bottom line price of the second interaction party. Wherein the bottom line price may be a preset value.

On the other hand, for the predictive equations used in the above process, in a simple manner, the parameters that the predictive equations may contain may be yield and cut values. The foregoing parameters may be understood as follows:

in the transaction process of each historical transaction object, the first interactive party carries out a first price cutting action, each historical transaction object has an initial bid, and the average value of a plurality of initial bids is the cut value in a predictive equation.

Firstly, a transaction curve is obtained according to the price cutting process of the interactive two parties on the historical transaction object, and optionally, the curve can be further subjected to smooth processing. The parameter value of the transaction curve is the yield ratio. For example, the trade curve is typically a power function, and the exponent of the power function is the yield.

In this embodiment, in the process of predicting the expected transaction action of the second interaction party, the transaction characteristic capable of reflecting the transaction habit of the first interaction party is used, so that the predicted expected transaction action more accords with the transaction habit of the first interaction party, and the transaction success rate is improved.

In summary, the three prediction models have different advantages and disadvantages, such as the model applicable to the embodiment shown in fig. 3a, which has high prediction accuracy, but requires a large number of training samples, and has slow convergence and high extractability. For example, the higher the extractability, the higher the profit value of the seller. In practical applications, the lower the extractability should be, the better. The MDP model has low prediction accuracy, few training samples, fast convergence and low extraction degree. The embodiment shown in fig. 3b is suitable for models with moderate prediction accuracy and moderate extractability. In practical application, the method can be selected according to the practical requirement.

In addition, the above method provides a plurality of prediction modes of expected transaction actions, and after the expected transaction actions are obtained, the response record can be generated in the mode of inquiring the statement template in the embodiment shown in fig. 1. Alternatively, the expected transaction actions may be input into the generative model by means of the generative model to output a response record through the generative model.

It should be noted that, in the above embodiments, the first interactive party may be considered as a buyer, and the second interactive party may be considered as a seller. However, in practical application, the two parties can also be exchanged, the first party can be considered as the seller, and the second party can be considered as the buyer, so that the buyer can realize different and personalized transactions for different sellers, and the success rate and the efficiency of the transactions are improved.

In addition to the transaction scenario described above, for a daily interactive scenario, the first interactive party may be a user and the second interactive party may be an intelligent robot.

The above embodiments describe the internal working process of the interactive device, but it is of course also possible to describe the interactive process from the interface point of view of the interactive device, i.e. from the point of view of the interactive party. Fig. 4 is a flowchart of another interaction method according to an embodiment of the present invention. The execution subject of the embodiment may also be an interactive device used by the interaction subject. As shown in fig. 4, the method may include the steps of:

S301, displaying an interactive interface comprising interactive buttons.

S302, responding to clicking operation triggered by the interaction main body on the interaction button, and acquiring interaction records aiming at target objects.

S303, determining expected interaction according to the interaction record and the identification information of the interaction main body.

S304, displaying a response record for describing the expected interaction in the interaction interface. In particular, the interaction agent may initiate a transaction for the target object through an interaction button provided by the interaction device in the interaction interface, optionally the interaction button corresponds to the interaction agent. And the interaction equipment responds to the clicking operation to acquire the interaction record of the interaction main body aiming at the target object. Meanwhile, the identification information of the interaction main body can be obtained, and the identification information reflects the transaction habit of the interaction main body. An expected interaction of the interaction agent is determined from the interaction record and the identification information. Finally, a response record is generated according to the expected interaction action, and the response record is also displayed on the interaction device of the interaction subject, namely, the automatic response to the interaction record is realized.

For details not described in detail in this embodiment, reference may be made to the related descriptions in the embodiments shown in fig. 1 to 3b, which are not described in detail herein.

In this embodiment, an interactive interface including an interactive button is displayed, and an interactive record of the interactive body on the target object is obtained in response to a click operation triggered by the interactive body on the button. And determining the expected interaction action according to the identification information of the interaction main body and the interaction record. Finally, the sentence to be recorded describing the desired interaction is displayed on the interaction device, i.e. the response to the interaction record. Therefore, the method is an efficient automatic interaction method, and meanwhile, when the response record is generated, the interaction equipment considers the identification information for reflecting the interaction habit of the interaction main body, so that the generated response record is more in line with the interaction habit of the interaction main body, and the transaction success rate is improved.

It should be noted that the interaction body may specifically include a first interaction party and a second interaction party, where the first interaction party is a purchaser of the target object, and the second interaction party is a seller of the target object. The embodiment shown in fig. 4 can now be understood as follows:

The first interactors can initiate transactions to the target object through the interaction device, and then interaction records aiming at the target object can be input. Then, the interactive device used by the second interactive party can receive the identification information of the first interactive party and the interactive record input by the first interactive party, and determine the expected interactive action of the second interactive party according to the identification information and the interactive record. As the interaction proceeds, the number of records generated by the interaction parties for the target object may increase, and the interaction parties may generate at least one record for the target object, where the at least one record includes the interaction record input by the first interaction party.

Finally, the interactive device can generate a response record according to the expected interaction action, and the response record is also sent to the interactive device used by the first interactive party, so that the response records are displayed on the interactive devices of the interactive parties, namely, the response of the second interactive party to the interactive record generated by the first interactive party is realized automatically.

The embodiments shown in fig. 1 to 4 can be considered that one of the two interaction parties is a manual online response, and the other interaction party is an automatic response of the interaction equipment. However, in practical application, especially in a transaction scene, both interaction parties can be automatically replied, so that the transaction efficiency can be improved to the greatest extent. Fig. 5 is a flowchart of another interaction method according to an embodiment of the present invention, as shown in fig. 5, the method may include the following steps:

S401, displaying an interactive interface comprising interactive buttons.

S402, in response to clicking operation triggered by the first interaction party on the interaction button, acquiring respective identification information of the first interaction party and the second interaction party and a first interaction record aiming at the target object.

Taking a transaction scenario as an example, a transaction APP may be installed on an interaction device used by the first interaction party, and an interaction button is further provided on an interaction interface of the transaction APP, where the interaction button corresponds to the first interaction party. The first interaction party can trigger clicking operation on the interaction button, and interaction equipment used by the first interaction party responds to the clicking operation to acquire respective interaction identifications of the first interaction party and the second interaction party and first interaction records of the interaction parties aiming at the target object at the current moment.

Wherein the interaction identification is used for reflecting the interaction habit of the interaction party. The content contained in the first interaction record is at least one interaction record generated by the interaction parties to the target object after the first interaction party triggers the clicking operation. The first interaction record describes the current interaction action of the second interaction party, and the interaction action can be, for example, the detailed information of the query target object, or the bid, etc. And the first interaction record is displayed on the interaction equipment used by the interaction parties at the same time.

S403, according to the first interaction record and the identification information of the second interaction party, determining the expected interaction action of the first interaction party.

The interaction device used by the first interactors may then predict an expected interaction of the first interactors based upon the interaction identification of the second interactors and the first interaction record.

An interaction device used in the first interaction mode can determine a first action prediction model of the first interaction party for the second interaction party according to identification information of the second interaction party. And inputting the first interaction record into a first action prediction model to predict the expected interaction action of the first interaction party through the first prediction model. The first prediction model may be any of the various prediction models provided in the above embodiments.

S404, displaying a first response record for describing the expected interaction action of the first interaction party.

The interactive device used in the first interactive mode may then also generate a first response record based on the intended interaction, the first response record being used to respond to the first interaction record in step 402. The generation of the first response record can be realized by means of a generation model, and the first response record is also displayed on the interaction device used by the interaction two parties.

S405, responding to click operation triggered by the second interaction party on the interaction button, and determining expected interaction action of the second interaction party according to the first interaction record, the first response record and the identification information of the first interaction party.

Furthermore, the interactive device used by the second interactive party may trigger a click operation on an interactive button provided on the interactive interface, where the interactive button corresponds to the second interactive party. The interaction device used by the second interaction party may predict an expected interaction of the second interaction party based on the identification information of the first interaction party, the first interaction record and the first response record.

An interaction device used in the second interaction mode can determine a second action prediction model of the second interaction party for the first interaction party according to the identification information of the first interaction party. And inputting the first interaction record and the first response record into a second action prediction model so as to predict the expected interaction action of the second interaction party through the second prediction model. The second prediction model may be a plurality of prediction models provided in the above embodiments. The same or different prediction models can be set in the interaction devices used by the interaction parties respectively.

And S406, displaying a second response record for describing the expected interaction action of the second interaction party.

Finally, the interactive device used in the second interactive mode may also generate a second response record according to the expected interaction, which record is used to respond to the first response record in step 404. Wherein the generation of the second answer record may be effected by means of a generation model. And the generated second response record is also displayed on the interactive equipment used by the interactive two parties.

In addition, the details not described in detail in this embodiment may be referred to the related descriptions in the embodiments shown in fig. 1 to 3b, which are not repeated here.

In this embodiment, after the first interaction party starts to interact, both interaction parties can automatically answer the interaction record of the other party, so that automatic transaction of both parties is realized, and transaction efficiency is ensured. And the response records output by the interaction equipment used by the interaction parties in the automatic reply process are in accordance with the transaction habit of the other party, so that the success rate of the transaction is ensured.

The interaction method provided above is schematically described below taking a transaction scenario as an example. The following can be understood in conjunction with fig. 6.

Assuming that the first interactive party is a buyer, the used interactive equipment is first terminal equipment, the second interactive party is a seller, the used interactive equipment is second terminal equipment, the target object is novel M, and the price is marked by 30 yuan.

The buyer can trigger a transaction starting operation on a transaction interface provided by the first terminal device, and input an interaction record describing transaction actions on the transaction interface: "novel in novel" in novel, the interaction record describes the interaction action of inquiring attribute information. The interaction records can be simultaneously displayed on terminal equipment of both buyers and sellers.

The second terminal device used by the seller responds to the start operation triggered by the buyer, so that the identification information of the buyer, such as the user ID and the attribute information of the novel M, is obtained, the action prediction model of the seller for the buyer is determined according to the identification information, and then the expected transaction action of the seller is output by utilizing the action prediction model. Finally, the interaction record describing the expected transaction action is automatically generated through the generation model as 'novel in the eighth aspect'. The generated interaction record can be displayed on the terminal equipment of the buyer and the seller. At this time, that is, automatic reply of the seller to the buyer is realized, and the two records can form a round of dialogue for the novel M.

The buyer may then proceed to enter an interaction record describing the slash action on the transaction interface, "20 yuan may be. At this time, the second terminal device may record the interaction: "novel" for novice "for eight new" and "can be used for 20-element" for input into the action prediction model together, so that the action prediction model outputs the expected transaction action of the seller as: if the cut price is not accepted, the generative model can generate an interactive record of 'devout yuan out, which is not out, namely an automatic response to the record of' 20 yuan can be. I.e. a further round of dialogue for novel M is implemented.

Automatic transaction of the seller to the buyer is achieved after at least one round of dialogue. The transaction efficiency can be improved by automatic transaction. The action prediction model is trained according to the historical transaction objects of the buyers and the transaction habit data of the buyers, so that the output is the content conforming to the transaction habit of the buyers, and the success rate of the transaction is ensured.

The interaction method provided above will be schematically described with reference to a daily chat scenario.

Assuming that the first interactive party is user a and the second interactive party is an intelligent robot. The target object may be object content discussed by both, such as a movie.

User a may send a voice command to the intelligent robot, i.e. an interactive recording of "what was most recently a good movie". The intelligent robot can determine the identification information of the user A through voiceprint recognition, and then determine the action prediction model corresponding to the user according to the identification information of the user A. The interaction record is input into the model to output a response record predicted by the mode action, namely' recently watched movies are many, you like comedy type movies most, XXX is good. At this time, the reply of the intelligent robot to the user A is realized.

Because the prediction model is trained according to the speaking habit of the user B, the output response record has the similar mood and thinking as the user B, so that the user A has a feeling of communicating with the user B, and the interactive experience of the user is ensured.

The following embodiments are described by taking a transaction scenario as an example. In this scenario, the interaction characteristic is a transaction characteristic. The historical object may be a commodity that the first interactors transact within a preset time period. And in the embodiments described below, the first party may be considered a buyer and the second party a seller.

Based on the foregoing, fig. 7 is a flowchart of a model training method according to an embodiment of the present invention, where the model training method according to the embodiment of the present invention may be performed by a training device. It will be appreciated that the training device may be implemented as software, or a combination of software and hardware, such as a server. As shown in fig. 7, the method includes the steps of:

S501, receiving an interaction record of a first interaction party corresponding to a history object, wherein the interaction record comprises a plurality of records generated by the first interaction party and a second interaction party, and the interaction record is input by the second interaction party.

S502, determining interaction characteristics of the first interaction party according to the interaction records.

The second interactors can determine which historical objects the first interactors have generated a transaction for within a preset time period, wherein the transaction comprises a successful transaction and a failed transaction. At the same time, the second interactive party may also collect at least one record, i.e. an interaction record, generated by the first interactive party and the second interactive party for these historical objects. In practical applications, different historical objects may correspond to different second interactors, but the last second interactors that collect the interaction records may be any second interactors on the transaction platform, and are not limited to the second interactors corresponding to the historical objects.

The second interactor may input the collected interaction record into a server. The server can also obtain the generation time of each record in the interaction records while receiving the interaction records, and the interaction characteristics of the first interaction party are obtained by analyzing and processing the interaction records.

In the transaction scenario, the interaction characteristic is a transaction characteristic, which may include at least one of bargained capability, transaction frequency, transaction efficiency, and transaction success rate of the first interaction party. For details, reference is made to the relevant description of the embodiment shown in fig. 3 a.

S503, determining action prediction parameters according to the interaction characteristics.

S504, outputting the motion prediction parameters so that the second interaction party obtains the prediction model parameters for the first interaction party according to the motion prediction parameters.

The server may further determine the motion prediction parameters according to the interaction characteristics of the first interaction party, and finally output the motion prediction parameters. For a second interaction party with model training requirements, the model parameters can be directly input into an established model structure, so that an action prediction model of the second interaction party relative to the first interaction party is obtained. The process of determining the parameters may be described in the following embodiments shown in fig. 9 to 11.

The embodiment can be considered as providing an online platform for model training, for a second interaction party with model training requirements, the second interaction party can input collected data to the platform so as to automatically output model parameters for the second interaction party, and the second interaction party can directly input the model parameters to a model so as to obtain the model, so that response to transaction records generated by the first interaction party is realized.

Fig. 8 is a flowchart of another model training method according to an embodiment of the present invention, where the model training method according to the embodiment of the present invention may be performed by a training device. It is understood that the training device may be implemented in software, or a combination of software and hardware. As shown in fig. 8, the method includes the steps of:

s601, an interaction record corresponding to the history object is obtained, wherein the interaction record comprises a plurality of record records generated by a first interaction party and a second interaction party.

S602, determining interaction characteristics of the first interaction party according to the interaction records.

S603, determining an action prediction model of the second interaction party for the first interaction party according to the interaction characteristics.

Taking a transaction scenario as an example, the interaction includes a transaction of the first interaction party and the second interaction party on the historical object within a preset time period. The interaction characteristics of the first interaction party include transaction characteristics of the first interaction party. The transaction characteristic comprises at least one of bargained capability, transaction frequency, transaction efficiency and transaction success rate of the first interaction party in a preset time period.

The steps of this embodiment are similar to those of the embodiment shown in fig. 7, and the specific implementation process may be referred to the related description in the embodiment shown in fig. 7, which is not repeated here.

In this embodiment, the interaction characteristics of the first interaction party are considered in the model training process, so that the second interaction party is trained to be specifically directed to the action prediction model of the first interaction party. The obtained model can automatically output the response record conforming to the interaction habit of the first interaction party, thereby ensuring the efficiency and success rate of the transaction.

In practical applications, the motion prediction model may be various, and the motion model may specifically be represented as an NDF model, and as shown in fig. 9, one determination manner of the motion prediction model, that is, an alternative implementation manner of step 603:

S701, determining utility loss parameters corresponding to the first interaction party according to the transaction characteristics of the first interaction party.

After the transaction characteristics are obtained, the data such as transaction frequency, transaction success rate, transaction efficiency, bargaining capacity and the like can be estimated with maximum likelihood, so that utility loss parameters are determined.

S702, determining a prediction equation and a utility equation according to the transaction characteristics.

For a description of the prediction equations and the utility equations, reference may be made to the relevant description in the embodiment shown in fig. 3 b. Two equations can be determined in the manner provided in the above embodiments, and in the process of equation parameter determination, the bargained ability in the transaction characteristics is emphasized.

S703, forming an action prediction model by the utility loss parameter, the utility equation and the prediction equation.

Finally, a formula for motion prediction, i.e. a motion prediction model, may be formed from the utility loss parameters, the utility equation and the prediction equation, as provided in the embodiment shown in fig. 3 b. The expected interaction of the second party is predicted by whether this determination formula holds.

In the training process provided in this embodiment, the transaction characteristics of the first interaction party are considered, so that it is ensured that the trained action prediction model is specific to the first interaction party by the second interaction party. In addition, what is not described in detail in this embodiment can be referred to as a related description in the embodiment shown in fig. 3 b.

As shown in fig. 10, another way to determine the motion prediction model is to implement step 603 as an alternative implementation:

S801, acquiring attribute information of a historical transaction object of a first interaction party and a plurality of transaction states described in an interaction record, wherein the transaction states comprise at least one transaction action described in the interaction record.

Attribute information of a historical transaction object of a first interactive party is obtained, and a plurality of transaction states described in an interaction record are obtained. Wherein the attribute information of the historical transaction object comprises the detail information of the historical transaction object and/or the transaction competitiveness of the historical transaction object. The transaction state may be considered as a sequence of actions made up of interactions described by each of the interaction records.

Receiving the transaction record shown in fig. 2, the transaction states included in the transaction record may be:

State 1 hacking price (first interaction party)

State 2-cut price (first Interactive) disagrees with cut price (second Interactive)

State 3 cut price (first interactors) -disagree with cut price (second interactors) -cut price (first interactors)

State 4 cut price (first interacting party) -disagree with cut price (second interacting party) -cut price (first interacting party) -disagree with cut price (second interacting party)

State 5-cut price (first interaction party) -disagree with cut price (second interaction party) -cut price (first interaction party)

Thus, each historical transaction object that it transacts may have multiple interaction states for the first interacting party.

S802, combining attribute information of historical transaction objects, determining probability values of expected transaction actions of a first interaction party as different transaction actions in a target transaction state, wherein the target transaction state is any one of a plurality of transaction states.

After obtaining the respective transaction states of the first interaction party aiming at different historical transaction objects, obtaining the probability values that the expected transaction actions of the first interaction party are different transaction actions, such as inquiring attribute information, cutting prices, agreeing to transaction actions and the like, in the target transaction state in a statistical mode. The expected transaction action corresponding to the highest probability value, namely the transaction action most frequently made by the first interaction party in the target transaction state, reflects the transaction habit of the first interaction party. Wherein, the target transaction state is any one of the transaction states in step 801.

S803, generating an action prediction model according to the transaction characteristics of the first interaction party, the probability values corresponding to the transaction states and the transaction characteristics of the first interaction party.

The mapping relation between different transaction states and probability values can be embodied as a Markov chain matrix, and the matrix can be regarded as parameters of the motion prediction model, namely the motion prediction model is obtained. This model may actually be an MDP model.

In the training process provided in this embodiment, the transaction characteristics of the first interaction party and the attribute information of the historical transaction object are considered at the same time, so that it is ensured that the trained action prediction model is specific to the first interaction party by the second interaction party.

It should be noted that, the motion prediction model generated above is specific to the first interaction party. However, in practical applications, it is not practical for the second interaction party to specially train a prediction model for each first interaction party, so after the prediction model is obtained in the above manner, the obtained interaction records of the first interaction party on the historical transaction objects can be further analyzed to determine the transaction characteristics of the first interaction party. The predictive model that has been obtained above applies not only to the first party but also to all parties having the same or similar transaction characteristics as the first party.

As shown in fig. 11, another way to determine the motion prediction model is an alternative implementation of step 603:

S901, acquiring attribute information of a historical transaction object of a first interaction party and an interaction record describing a plurality of transaction states, wherein the transaction states comprise at least one transaction action described in the interaction record.

S902, an action prediction model is input to train the action prediction model, where the target transaction state is any one of a plurality of transaction states, and the interaction record of the expected transaction action of the second interaction party in the target state, the transaction characteristic of the first interaction party, and the attribute information of the historical transaction object are described.

Attribute information of a historical transaction object of a first interactive party is obtained, and an interaction record generated for the historical transaction object is obtained. The interaction record describes a plurality of transaction states aiming at the historical transaction objects, and specifically, the interaction record can describe transaction actions of a first interaction party in a target state and expected transaction actions of a second interaction party in the target state. The attribute information of the historical transaction object includes detail information of the historical transaction object and/or transaction competitiveness of the historical transaction object.

And then, training the training equipment according to the acquired interaction record, the transaction characteristics of the first interaction party and the attribute information of the historical transaction object as training samples, thereby obtaining the action prediction model. Wherein the motion prediction model may be an RNN model.

The action prediction model outputs an expected transaction action of the second party, which may be considered as a predicted transaction action, for the input interaction record describing the current interaction action of the first party. And since the interaction record entered into the model already contains the expected action of the second interaction party, which is a real transaction action, optionally a semantic similarity between the predicted transaction action and the real transaction action can be calculated. The higher the semantic similarity between the two, the better the training effect of the model. The similarity value may be indirectly considered as a loss value of the model, and thus the model parameters may be adjusted according to this similarity.

In another alternative mode of model parameter adjustment, the action prediction model outputs that the predicted transaction action is an expected transaction action of the second interaction party in the target state, and if the predicted transaction action does not meet a preset transaction rule with at least one transaction action included in the target transaction state described in the interaction record, the output predicted transaction action is indicated to be incorrect and does not meet the transaction logic. At this time, a large loss value may be set for this output so as to adjust model parameters of the motion prediction model according to the loss value.

The preset transaction rule comprises rule 1 that the first interactive party bid monotonically increases and the second interactive party bid monotonically decreases.

Rule 2 the second interactive bid is higher than the first interactive bid.

And 3, the price of the achievement is larger than or equal to the last round of bidding of the first interaction party and smaller than or equal to the last round of bidding of the second interaction party.

Based on the above rules, the target transaction state is, for example, 100 yuan for the original bid (second interactors) -cut to 80 yuan (first interactors). At this point, the second party's actual expected transaction actions are not agreeing to cut prices, and are re-bid for 90 yuan, as known from the interaction records that have been obtained. If the predictive action entered by the predictive model is not agreed to cut prices, 60-membered is re-bid, which is obviously not in line with the trading logic.

In fact, if the model outputs the expected transaction action which does not conform to the transaction logic, the model is punished, namely a larger loss value is set.

In addition, if the expected transaction action output by the model accords with the transaction logic and can maximize the benefit of the first interaction party, the first interaction party is rewarded, and a rewarding value is obtained to offset the loss value.

Alternatively, the prize value ：γ_s＝p_T-υ_s+ω_d*II_d,γ_b＝υ_b-p_T+ω_d*II_d. may be calculated by the following equation, where γ _s is the prize value of the second interaction party, γ _b is the prize value of the first interaction party, p _T is the proportion of the final transaction, ω _d is the specific gravity of the transaction benefit, if the transaction is successful, ω _d =1, otherwise ω _d＝0,II_d is the indication function to indicate whether the transaction is present or not, v _s is the psychological price of the second interaction party, v _b is the psychological price of the first interaction party, and the psychological price may be considered as the lowest price that the interaction party can receive, which may be preset.

In addition, the embodiments shown in fig. 9 to 11 actually train a first motion prediction model of the second interaction party with respect to the first interaction party. In practical applications, the second motion prediction model of the first interaction party with respect to the second interaction party can be trained in the same way as described above. At this time, in order to further ensure the training effect of the model, the two motion prediction models may be optionally further subjected to unsupervised training.

Specifically, the first motion prediction model is deployed in an interactive device used by the second interactive party, and the second motion prediction model is deployed in an interactive device used by the first interactive party. For the interaction records output by one action prediction model, the other action prediction model can automatically answer the interaction records, namely the interaction records are generated. The two models can further finish the training of the models according to the interaction records automatically output by the models and automatically output by the other party, namely the training effect of the models is guaranteed.

The interaction means of one or more embodiments of the present invention will be described in detail below. Those skilled in the art will appreciate that these image replacement devices can be configured by the steps taught by the present solution using commercially available hardware components.

Fig. 12 is a schematic structural diagram of an interaction device according to an embodiment of the present invention, as shown in fig. 12, where the interaction device includes:

the obtaining module 11 is configured to obtain identification information of an interaction subject and an interaction record for a target object, where the interaction record is used to describe an interaction action.

An action determining module 12 for determining an expected interaction action based on the identification information and the interaction record.

And the generating module 13 is used for generating a response record corresponding to the interaction record according to the expected interaction action.

Optionally, the interaction comprises a transaction action of the interaction body on the target object. The transaction action includes any one of a cut-and-price action on the target object, a query action on attribute information of the target object, a transaction success action on the target object, and a transaction failure action on the target object. The interaction body includes a first interaction party and a second interaction party.

The action determining module 12 is specifically configured to determine, according to the identification information of the first interactive party and the interaction record describing the current interaction action of the first interactive party, an expected transaction action of the second interactive party.

Optionally, the action determining module 12 specifically includes:

A determining unit 121, configured to determine, according to the identification information of the first interaction party, a motion prediction model of the second interaction party with respect to the first interaction party.

An input module 122 for inputting the interaction record describing the current transaction action of the first interaction party into the action prediction model to output the expected transaction action of the second interaction party through the action prediction model.

Optionally, the input module 122 is specifically configured to input the interaction record, attribute information of the target object, and transaction characteristics of the first interaction party and the second interaction party into the motion prediction model, so as to perform feature extraction on the interaction record, the attribute information, and the transaction characteristics through a feature extraction layer of the motion prediction model;

And inputting the extracted characteristics into an output layer of the action prediction model to output the expected transaction action of the second interaction party through the output layer.

The transaction characteristics comprise at least one of bargained capability, transaction frequency, transaction efficiency and transaction success rate of the first interaction party in a preset time period, the attribute information of the historical transaction object comprises detail information of the historical transaction object and/or transaction competitiveness of the historical transaction object, and the attribute information of the target object comprises detail information of the target transaction object and/or transaction competitiveness of the target transaction object.

Optionally, the input module 122 is specifically configured to input the interaction record and attribute information of the target object into the action prediction model, so as to output, through the action prediction model, an expected transaction action of the second interaction party, where model parameters of the action prediction model are determined according to the transaction characteristics of the first interaction party and attribute information of the historical transaction object of the first interaction party.

Optionally, the input module 122 is specifically configured to obtain a current transaction action of the first interaction party described in the interaction record;

and determining the expected transaction action of the second interaction party according to the current transaction action and a prediction equation in the action prediction model, wherein the prediction equation is determined according to the transaction characteristic of the first interaction party.

Optionally, the input module 122 is specifically configured to determine that the candidate transaction action is the expected transaction action if a preset condition is satisfied between the current transaction action of the first interaction party and the candidate transaction action of the second interaction party output by the action prediction model.

Optionally, the generating module 13 is further configured to generate a response record indicating that the transaction is successful if a preset condition is not satisfied between the current transaction action and the alternative transaction action.

Optionally, the apparatus further comprises:

The utility determining module 14 is configured to determine, according to a utility equation, respective utility values of the current transaction action and the candidate transaction action, where the utility value of the current transaction action reflects a satisfaction degree of the first interaction party after the target object is transacted with the current transaction action.

The judging module 15 is configured to determine whether the current transaction action and the alternative transaction action meet the preset condition according to the utility value and a utility loss parameter, where the utility loss parameter is determined according to the transaction characteristic of the first interaction party.

Optionally, the generating module 13 is further configured to input the expected transaction actions into the generating model, so as to output the response record through the generating model.

The device of fig. 12 may perform the method of the embodiment of fig. 1 to 3b, and reference is made to the relevant description of the embodiment of fig. 1 to 3b for parts of this embodiment not described in detail. The implementation process and technical effects of this technical solution are described in the embodiments shown in fig. 1 to 3b, and are not described herein.

The internal functions and structures of the interaction means are described above, and in one possible design the structure of the interaction means may be implemented as an electronic device, as shown in fig. 13, which may comprise a processor 21 and a memory 22. Wherein the memory 22 is for storing a program for supporting the electronic device to perform the interaction method provided in the embodiments shown in fig. 1-3 b described above, and the processor 21 is configured for executing the program stored in the memory 22.

The program comprises one or more computer instructions which, when executed by the processor 21, are capable of carrying out the steps of:

Optionally, the processor 21 is further configured to perform all or part of the steps in the embodiments shown in fig. 1 to 3 b.

The structure of the electronic device may further include a communication interface 23, for the electronic device to communicate with other devices or a communication network.

In addition, an embodiment of the present invention provides a computer storage medium, configured to store computer software instructions for the electronic device, where the computer storage medium includes a program for executing the interaction method according to the embodiment of the method shown in fig. 1 to 3 b.

Fig. 14 is a schematic structural diagram of another interaction device according to an embodiment of the present invention, as shown in fig. 14, where the interaction device includes:

And a display module 31 for displaying an interactive interface including interactive buttons, wherein the interactive buttons correspond to the interactive main body.

And the obtaining module 32 is configured to obtain an interaction record for the target object in response to a click operation triggered by the interaction button by the interaction body.

The action determining module 32 is configured to determine an expected interaction action according to the interaction record and the identification information of the interaction body.

The display module 31 is further configured to display a response record for describing the expected interaction in the interaction interface.

The apparatus shown in fig. 14 may perform the method of the embodiment shown in fig. 4, and reference is made to the relevant description of the embodiment shown in fig. 4 for parts of this embodiment not described in detail. The implementation process and the technical effect of this technical solution are described in the embodiment shown in fig. 4, and are not described herein.

The internal functions and structures of the interaction means are described above, and in one possible design the structure of the interaction means may be implemented as an electronic device, as shown in fig. 15, which may comprise a processor 41 and a memory 42. Wherein the memory 42 is for storing a program for supporting the electronic device to execute the interaction method provided in the embodiment shown in fig. 4 described above, and the processor 41 is configured for executing the program stored in the memory 42.

The program comprises one or more computer instructions which, when executed by the processor 41, are capable of carrying out the steps of:

displaying an interactive interface comprising interactive buttons;

Optionally, the processor 41 is further configured to perform all or part of the steps in the embodiment shown in fig. 4.

The electronic device may further include a communication interface 43 for the electronic device to communicate with other devices or a communication network.

In addition, an embodiment of the present invention provides a computer storage medium, configured to store computer software instructions for the electronic device, where the computer storage medium includes a program for executing the interaction method in the embodiment of the method shown in fig. 4.

Fig. 16 is a schematic structural diagram of another interaction device according to an embodiment of the present invention, as shown in fig. 16, where the interaction device includes:

And a display module 51 for displaying an interactive interface including interactive buttons.

And the obtaining module 52 is configured to obtain, in response to a click operation triggered by the first interaction party on the interaction button, identification information of each of the first interaction party and the second interaction party, and a first interaction record for the target object.

A first action determining module 53, configured to determine an expected interaction action of the first interaction party according to the first interaction record and the identification information of the second interaction party.

The display module 51 is further configured to display a first response record describing an expected interaction of the first interaction party.

And a second action determining module 54, configured to determine, in response to a click operation triggered by the second interaction party on the interaction button, an expected interaction action of the second interaction party according to the first interaction record, the first response record, and the identification information of the first interaction party.

The display module 51 is further configured to display a second answer record describing the expected interaction of the second interaction party.

Optionally, the first action determining module 52 is specifically configured to determine, according to the identification information of the second interaction party, a first action prediction model of the first interaction party with respect to the second interaction party;

a first interaction record describing a current interaction of the second interaction party is input into the first action prediction model to output an expected interaction of the first interaction party through the first action prediction model.

Optionally, the second action determining module 54 is specifically configured to determine, according to the identification information of the first interaction party, a second action prediction model of the second interaction party with respect to the first interaction party;

a first interaction record describing a current interaction of the second interaction party and the first response record describing an expected interaction of the first interaction party are input into the second action prediction model to output the expected interaction of the second interaction party through the second action prediction model.

The apparatus shown in fig. 16 may perform the method of the embodiment shown in fig. 5, and reference is made to the relevant description of the embodiment shown in fig. 5 for parts of this embodiment not described in detail. The implementation process and the technical effect of this technical solution are described in the embodiment shown in fig. 5, and are not described herein.

The internal functions and structures of the interaction means are described above, and in one possible design the structure of the interaction means may be implemented as an electronic device, as shown in fig. 17, which may comprise a processor 61 and a memory 62. Wherein the memory 62 is for storing a program for supporting the electronic device to execute the interaction method provided in the embodiment shown in fig. 5 described above, and the processor 61 is configured for executing the program stored in the memory 62.

The program comprises one or more computer instructions which, when executed by the processor 61, are capable of carrying out the steps of:

displaying an interactive interface comprising interactive buttons;

Optionally, the processor 61 is further configured to perform all or part of the steps in the embodiment shown in fig. 5.

The electronic device may further include a communication interface 63 in the structure of the electronic device, for communicating with other devices or a communication network.

In addition, an embodiment of the present invention provides a computer storage medium, configured to store computer software instructions for the electronic device, where the computer storage medium includes a program for executing the interaction method in the embodiment of the method shown in fig. 5.

Fig. 18 is a schematic structural diagram of a model training device according to an embodiment of the present invention, as shown in fig. 18, where the device includes:

The receiving module 71 is configured to receive an interaction record corresponding to a history object by a first interaction party, where the interaction record includes a plurality of records generated by the first interaction party and a second interaction party, and the interaction record is input by the second interaction party.

A characteristic determination module 72 is configured to determine an interaction characteristic of the first interaction party according to the interaction record.

A parameter determination module 73 for determining an action prediction parameter based on the interaction characteristics.

An output module 74 for outputting the motion prediction parameters to enable the second interaction party to obtain prediction model parameters for the first interaction party according to the motion prediction parameters.

The apparatus shown in fig. 18 may perform the method of the embodiment shown in fig. 7, and reference is made to the relevant description of the embodiment shown in fig. 7 for parts of this embodiment not described in detail. The implementation process and the technical effect of this technical solution are described in the embodiment shown in fig. 7, and are not described herein.

The internal functions and structures of the model training apparatus are described above, and in one possible design, the structure of the interaction means may be implemented as an electronic device, which may include a processor 81 and a memory 82, as shown in FIG. 19. Wherein the memory 82 is configured to store a program for supporting the electronic device to execute the model training method provided in the embodiment shown in fig. 7 described above, and the processor 81 is configured to execute the program stored in the memory 82.

The program comprises one or more computer instructions which, when executed by the processor 81, are capable of carrying out the steps of:

Optionally, the processor 81 is further configured to perform all or part of the steps in the embodiment shown in fig. 7.

The electronic device may further include a communication interface 53 in the structure of the electronic device, for the electronic device to communicate with other devices or a communication network.

In addition, an embodiment of the present invention provides a computer storage medium for storing computer software instructions for use in the above-described electronic device, where the computer storage medium includes a program for executing the model training method according to the embodiment of the method shown in fig. 7.

Fig. 20 is a schematic structural diagram of another model training apparatus according to an embodiment of the present invention, as shown in fig. 20, where the apparatus includes:

an obtaining module 91, configured to obtain an interaction record corresponding to the history object, where the interaction record includes a plurality of records generated by the first interaction party and the second interaction party.

A characteristic determining module 92, configured to determine an interaction characteristic of the first interaction party according to the interaction record.

A model determining module 93, configured to determine, according to the interaction characteristics, a motion prediction model of the second interaction party with respect to the first interaction party.

Optionally, the interaction comprises a transaction of the first interaction party and the second interaction party on the historical object within a preset time period, the interaction characteristic of the first interaction party comprises a transaction characteristic of the first interaction party, and the transaction characteristic comprises at least one of bargained capability, transaction frequency, transaction efficiency and transaction success rate of the first interaction party within the preset time period.

The model determining module 93 is configured to determine a utility loss parameter corresponding to the first interaction party according to a transaction characteristic of the first interaction party;

determining a prediction equation and a utility equation according to the transaction characteristics;

the action prediction model is formed by the utility loss parameter, the utility equation and the prediction equation.

Optionally, the attribute information of the historical transaction object includes detail information of the historical transaction object and/or transaction competitiveness of the historical transaction object.

The model determining module 93 is configured to obtain attribute information of a historical transaction object of the first interaction party and a plurality of transaction states described in the interaction record, where the transaction states include at least one transaction action described in the interaction record;

Determining the probability value of the expected transaction action of the first interaction party as different transaction actions in the target transaction state according to the attribute information of the historical transaction object, wherein the target transaction state is any transaction state in the plurality of transaction states;

And generating the action prediction model according to the transaction characteristics of the first interaction party and the probability values corresponding to the transaction states.

Optionally, the model determining module 93 is configured to obtain attribute information of a historical transaction object of the first interaction party and the interaction record describing a plurality of transaction states, where a transaction state includes at least one transaction action described in the interaction record;

An interaction record of the target transaction state, the expected transaction action of the second interaction party in the target state, the transaction characteristic of the first interaction party and the attribute information of the historical transaction object are described, and an action prediction model is input to train the action prediction model, wherein the target transaction state is any transaction state in the transaction states.

Optionally, the apparatus further comprises an adjustment module 94 for adjusting model parameters of the predictive model according to semantic similarity between the predicted transaction actions output by the action predictive model and the expected transaction actions of the second interacting party.

Optionally, the adjusting module 94 is further configured to adjust a model parameter of the prediction model if a preset transaction rule is not satisfied between a predicted transaction action output by the action prediction model and at least one transaction action included in the target transaction state, where the at least one transaction action is generated by the first transaction party and/or the second transaction party.

Optionally, the device further comprises an output module 95, configured to respond to a transaction start operation, so that the first interaction party directs to a first action prediction model of the second interaction party, and the second interaction party directs to a second action prediction model of the first interaction party, and alternately output response records corresponding to the interaction records according to interaction records generated by the other party.

The adjustment module 94 is further configured to adjust model parameters of each of the first motion prediction model and the second motion prediction model according to the response record.

The apparatus shown in fig. 20 may perform the method of the embodiment shown in fig. 8, and reference is made to the relevant description of the embodiment shown in fig. 8 for parts of this embodiment not described in detail. The implementation process and the technical effect of this technical solution refer to the description in the embodiment shown in fig. 8, and are not repeated here.

The internal functions and structures of the model training apparatus are described above, and in one possible design, the structure of the interaction means may be implemented as an electronic device, as shown in fig. 21, which may include a processor 101 and a memory 102. Wherein the memory 102 is for storing a program for supporting the electronic device to execute the model training method provided in the embodiment shown in fig. 8 described above, and the processor 101 is configured for executing the program stored in the memory 52.

The program comprises one or more computer instructions, wherein the one or more computer instructions, when executed by the processor 101, are capable of performing the steps of:

Optionally, the processor 101 is further configured to perform all or part of the steps in the foregoing embodiment shown in fig. 8.

The structure of the electronic device may further include a communication interface 103, which is used for the electronic device to communicate with other devices or a communication network.

In addition, an embodiment of the present invention provides a computer storage medium for storing computer software instructions for use in the above-described electronic device, where the computer storage medium includes a program for executing the model training method according to the embodiment of the method shown in fig. 8.

It should be noted that the above-mentioned embodiments are merely for illustrating the technical solution of the present invention, and not for limiting the same, and although the present invention has been described in detail with reference to the above-mentioned embodiments, it should be understood by those skilled in the art that the technical solution described in the above-mentioned embodiments may be modified or some technical features may be equivalently replaced, and these modifications or substitutions do not make the essence of the corresponding technical solution deviate from the spirit and scope of the technical solution of the embodiments of the present invention.

Claims

1. An interaction method, comprising:

acquiring identification information of an interaction main body and an interaction record aiming at a target object, wherein the interaction record is used for describing interaction actions, the interaction actions comprise transaction actions of the interaction main body on the target object, and the interaction main body comprises a first interaction party and a second interaction party;

The method comprises the steps of determining an action prediction model of a second interaction party for a first interaction party according to identification information of the first interaction party, inputting the interaction record describing current transaction actions of the first interaction party into the action prediction model to output expected transaction actions of the second interaction party through the action prediction model, inputting the interaction record, attribute information of a target object and transaction characteristics of the first interaction party and the second interaction party into the action prediction model to obtain expected transaction actions of the second interaction party, or inputting the interaction record and the attribute information of the target object into the action prediction model to obtain expected transaction actions of the second interaction party, and generating response records corresponding to the interaction records according to the expected transaction actions.

2. The method of claim 1, wherein the transaction action comprises any one of a slashing action on the target object, a query action on attribute information of the target object, a transaction success action on the target object, and a transaction failure action on the target object.

3. The method of claim 1, wherein the interaction record describing the current interaction of the first party, inputting the action prediction model to output the expected transaction actions of the second party via the action prediction model, comprises:

inputting the interaction record, the attribute information of the target object and the transaction characteristics of the first interaction party and the second interaction party into the action prediction model so as to perform feature extraction on the interaction record, the attribute information and the transaction characteristics through a feature extraction layer of the action prediction model;

4. The method of claim 1, wherein the interaction record describing the current interaction of the first party, inputting the action prediction model to output the expected transaction actions of the second party via the action prediction model, comprises:

And inputting the interaction records and the attribute information of the target object into the action prediction model to output the expected transaction action of the second interaction party through the action prediction model, wherein the model parameters of the action prediction model are determined according to the transaction characteristics of the first interaction party and the attribute information of the historical transaction object of the first interaction party.

5. The method of claim 1, wherein the interaction record describing the current interaction of the first party, inputting the action prediction model to output the expected transaction actions of the second party via the action prediction model, comprises:

acquiring a current transaction action of the first interaction party described in the interaction record;

6. The method of claim 5, wherein said determining the expected transaction actions of the second party based on the current transaction actions and predictive equations in the action predictive model comprises:

And if the current transaction action of the first interaction party and the alternative transaction action of the second interaction party output by the action prediction model meet the preset condition, determining the alternative transaction action as the expected transaction action.

7. The method of claim 6, wherein the method further comprises:

and if the current transaction action and the alternative transaction action do not meet the preset condition, generating a response record indicating successful transaction.

8. The method of claim 6, wherein the method further comprises:

Determining respective utility values of the current transaction action and the alternative transaction action according to a utility equation, wherein the utility value of the current transaction action reflects the satisfaction degree of the first interaction party after the target object is intersected with the current transaction action;

and determining whether the current transaction action and the alternative transaction action meet the preset condition or not according to the utility value and the utility loss parameter, wherein the utility loss parameter is determined according to the transaction characteristic of the first interaction party.

9. The method according to any one of claims 1 to 8, wherein generating a response record corresponding to the interaction record according to the expected transaction action comprises:

Inputting the expected transaction actions into a generation model to output the response records through the generation model.

10. The method according to any one of claims 3 to 8, wherein the transaction characteristic comprises at least one of bargained capability, transaction frequency, transaction efficiency, transaction success rate of the first interactive party, the attribute information of the historical transaction object comprises details of the historical transaction object and/or transaction competitiveness of the historical transaction object, and the attribute information of the target object comprises details of the target transaction object and/or transaction competitiveness of the target transaction object within a preset time period.

11. An interaction method, comprising:

displaying an interactive interface comprising interactive buttons;

The method comprises the steps of responding to click operation triggered by an interaction main body on an interaction button, acquiring an interaction record aiming at a target object, wherein the interaction action comprises transaction action of the interaction main body on the target object, and the interaction main body comprises a first interaction party and a second interaction party;

The method comprises the steps of determining an action prediction model of a second interaction party for a first interaction party according to identification information of the first interaction party, inputting the interaction record describing current transaction actions of the first interaction party into the action prediction model to output expected transaction actions of the second interaction party through the action prediction model, inputting attribute information of the interaction record and the target object and respective transaction characteristics of the first interaction party and the second interaction party into the action prediction model to obtain expected transaction actions of the second interaction party, or inputting attribute information of the interaction record and the target object into the action prediction model to obtain expected transaction actions of the second interaction party;

and displaying a response record for describing the expected transaction action in the interactive interface.

12. The method of claim 11, wherein the interactive button corresponds to an interactive body.

13. An interaction method, comprising:

displaying an interactive interface comprising interactive buttons;

14. The method of claim 13, wherein determining the intended interaction of the first party based on the first interaction record and the identification information of the second party comprises:

Determining a first action prediction model of the first interaction party for the second interaction party according to the identification information of the second interaction party;

15. The method of claim 13, wherein the determining the intended interaction of the second party based on the first interaction record, the first response record, and the identification information of the first party comprises:

determining a second action prediction model of the second interaction party for the first interaction party according to the identification information of the first interaction party;

16. A method of model training, comprising;

17. A method of model training, comprising:

And determining an action prediction model of the second interaction party for the first interaction party according to the interaction characteristics, wherein the action prediction model is used for outputting the expected transaction action of the second interaction party according to the interaction record describing the current transaction action of the first interaction party, the attribute information of the target object and the transaction characteristics of the first interaction party and the second interaction party, or outputting the expected transaction action of the second interaction party according to the interaction record describing the current transaction action of the first interaction party and the attribute information of the target object.

18. The method of claim 17, wherein interacting comprises a transaction of the first and second interactors with the historical object for a preset time period, wherein the interaction characteristic of the first interactors comprises a transaction characteristic of the first interactors, and wherein the transaction characteristic comprises at least one of bargained capability, transaction frequency, transaction efficiency and transaction success rate of the first interactors for the preset time period.

19. The method of claim 18, wherein the determining an action prediction model of the second interaction party for the first interaction party based on the interaction characteristics comprises:

determining utility loss parameters corresponding to the first interaction party according to the transaction characteristics of the first interaction party;

20. The method of claim 18, wherein the determining an action prediction model of the second interaction party for the first interaction party based on the interaction characteristics comprises:

acquiring attribute information of a historical transaction object of the first interactive party and a plurality of transaction states described in the interaction record, wherein the transaction states comprise at least one transaction action described in the interaction record;

determining the probability value of the expected transaction action of the first interaction party as different transaction actions in a target transaction state according to the attribute information of the historical transaction object, wherein the target transaction state is any transaction state in the plurality of transaction states;

21. The method of claim 18, wherein the determining an action prediction model of the second interaction party for the first interaction party based on the interaction characteristics comprises:

Acquiring attribute information of a historical transaction object of the first interaction party and the interaction record describing a plurality of transaction states, wherein the transaction states comprise at least one transaction action described in the interaction record;

An action prediction model is input to train the action prediction model, wherein the target transaction state is any one of the transaction states, and the interaction record of the expected transaction action of the second interaction party, the transaction characteristic of the first interaction party and the attribute information of the historical transaction object are described in the target transaction state.

22. The method according to claim 20 or 21, wherein the attribute information of the historical transaction object comprises details information of the historical transaction object and/or transaction competitiveness of the historical transaction object.

23. The method of claim 21, wherein the method further comprises:

And adjusting model parameters of the prediction model according to semantic similarity between the predicted transaction action output by the action prediction model and the expected transaction action of the second interaction party.

24. The method of claim 21, wherein the method further comprises:

And if the predicted transaction action output by the action prediction model and at least one transaction action contained in the target transaction state do not meet a preset transaction rule, adjusting model parameters of the prediction model, wherein the at least one transaction action is generated by a first transaction party and/or a second transaction party.

25. The method of claim 21, wherein the method further comprises:

Responding to a transaction starting operation, so that a first action prediction model of the first interaction party aiming at the second interaction party and a second action prediction model of the second interaction party aiming at the first interaction party alternately output response records corresponding to interaction records generated by the other party;

And adjusting respective model parameters of the first motion prediction model and the second motion prediction model according to the response record.

26. An interactive apparatus, comprising:

The system comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring identification information of an interaction main body and an interaction record aiming at a target object, the interaction record is used for describing interaction actions, the interaction actions comprise transaction actions of the interaction main body on the target object, and the interaction main body comprises a first interaction party and a second interaction party;

The system comprises an action determining module, a generation module and a response record generating module, wherein the action determining module is used for determining an action prediction model of the second interaction party for the first interaction party according to identification information of the first interaction party, inputting the interaction record describing current transaction action of the first interaction party into the action prediction model to output expected transaction action of the second interaction party through the action prediction model, the action prediction model is input to obtain expected transaction action of the second interaction party according to the interaction record, attribute information of the target object and transaction characteristics of the first interaction party and the second interaction party, or the interaction record and attribute information of the target object are input into the action prediction model to obtain expected transaction action of the second interaction party, and the generation module is used for generating the response record corresponding to the interaction record according to the expected transaction action.

27. An electronic device comprising a memory, a processor, wherein the memory has executable code stored thereon that, when executed by the processor, causes the processor to perform the interaction method of any of claims 1-10.

28. A non-transitory machine-readable storage medium having stored thereon executable code, which when executed by a processor of an electronic device, causes the processor to perform the interaction method of any of claims 1-10.

29. An interactive apparatus, comprising:

The system comprises an acquisition module, an interaction module and a target object, wherein the acquisition module is used for responding to click operation triggered by an interaction main body on the interaction button and acquiring interaction records aiming at the target object;

The action determining module is used for determining an action prediction model of the second interaction party for the first interaction party according to the identification information of the first interaction party, inputting the interaction record describing the current transaction action of the first interaction party into the action prediction model to output the expected transaction action of the second interaction party through the action prediction model, wherein the interaction record, the attribute information of the target object and the transaction characteristics of the first interaction party and the second interaction party are input into the action prediction model to obtain the expected transaction action of the second interaction party, or the attribute information of the interaction record and the target object are input into the action prediction model to obtain the expected transaction action of the second interaction party;

the display module is further used for displaying a response record for describing the expected transaction action in the interactive interface.

30. An electronic device comprising a memory, a processor, wherein the memory has executable code stored thereon that, when executed by the processor, causes the processor to perform the interaction method of claim 11 or 12.

31. A non-transitory machine-readable storage medium having stored thereon executable code which, when executed by a processor of an electronic device, causes the processor to perform the interaction method of claim 11 or 12.

32. An interactive apparatus, comprising:

the acquisition module is used for responding to clicking operation triggered by the first interaction party on the interaction button and acquiring respective identification information of the first interaction party and the second interaction party and a first interaction record aiming at a target object;

33. An electronic device comprising a memory, a processor, wherein the memory has executable code stored thereon that, when executed by the processor, causes the processor to perform the interaction method of any of claims 13-15.

34. A non-transitory machine-readable storage medium having stored thereon executable code, which when executed by a processor of an electronic device, causes the processor to perform the interaction method of any of claims 13-15.

35. A model training device, comprising:

36. An electronic device comprising a memory, a processor, wherein the memory has executable code stored thereon that, when executed by the processor, causes the processor to perform the model training method of claim 16.

37. A non-transitory machine-readable storage medium having executable code stored thereon, which when executed by a processor of an electronic device, causes the processor to perform the model training method of claim 16.

38. A model training device, comprising:

The model determining module is used for determining an action prediction model of the second interaction party for the first interaction party according to the interaction characteristics;

The action prediction model is used for outputting the expected transaction action of the second interaction party according to the interaction record describing the current transaction action of the first interaction party, the attribute information of the target object and the transaction characteristics of the first interaction party and the second interaction party, or is used for outputting the expected transaction action of the second interaction party according to the interaction record describing the current transaction action of the first interaction party and the attribute information of the target object.

39. An electronic device comprising a memory, a processor, wherein the memory has executable code stored thereon that, when executed by the processor, causes the processor to perform the model training method of any of claims 17-25.

40. A non-transitory machine-readable storage medium having stored thereon executable code, which when executed by a processor of an electronic device, causes the processor to perform the model training method of any of claims 17 to 25.