CN115017285B

CN115017285B - Session response method, medium, device and computing equipment

Info

Publication number: CN115017285B
Application number: CN202210617493.8A
Authority: CN
Inventors: 徐梓钧; 冯旻伟; 尹竞成; 阮良
Original assignee: Hangzhou Netease Zhiqi Technology Co Ltd
Current assignee: Hangzhou Netease Zhiqi Technology Co Ltd
Priority date: 2022-06-01
Filing date: 2022-06-01
Publication date: 2025-08-01
Anticipated expiration: 2042-06-01
Also published as: CN115017285A

Abstract

The embodiments of the present disclosure provide a conversation answering method, medium, apparatus, and computing device, wherein the conversation answering method includes: obtaining a user's first question, and then, based on the first question, determining a target agent, obtaining the target agent's answering style based on the first question and the target agent's identity, and thereby, generating a corresponding answer based on the answering style and the first question. The method of the present disclosure considers the agent's answering style so that the subsequently generated answer is consistent with the agent's language style, thereby improving user experience and service quality. It also reduces the time it takes for a human agent to modify the answer content in order to respond, thereby improving service efficiency. In addition, the embodiments of the present disclosure provide a computing device.

Description

Session response method, medium, device and computing equipment

Technical Field

Embodiments of the present disclosure relate to the field of computer technology, and more particularly, to a session response method, medium, apparatus, and computing device.

Background

This section is intended to provide a background or context to the embodiments of the disclosure recited in the claims. The description herein is not admitted to be prior art by inclusion in this section.

With the continuous development of technology, the traditional business development is kept by increasing the number of people to encounter a bottleneck, and intelligent customer service interaction technology is generated gradually to replace traditional manual service, so that the importance of efficient interaction between users and intelligent customer service to enterprises is self-evident.

In the related art, intelligent customer service can only be used for replying a few simple repeated questions, and aiming at the phenomena of large matching error and low reply accuracy when more complex questions exist, a manual seat is required to be switched to reply correspondingly. However, the manual seat is relatively short, and needs to be combined with a customer service system, for example, the customer service system generates a response based on the user's problem, and the manual seat refers to the response to answer the problem presented by the user.

However, when the manual agent refers to the response generated by the customer service system and answers the questions presented by the user, the language style is often changed, so that the user experience is poor and the service quality is affected.

Disclosure of Invention

The invention provides a session response method, a session response medium, a session response device and a session response computing device, so that a response generated by a customer service system is consistent with the language style of a seat, and user experience is improved.

In a first aspect of the embodiments of the present disclosure, a session answer method is provided, including obtaining a first question, determining a target agent according to the first question, obtaining a reply style of the target agent based on the first question and an identity of the target agent, and generating reply content according to the reply style and the first question.

In one embodiment of the disclosure, before generating the reply content according to the reply style and the first question, the method further comprises determining a topic vector corresponding to the first question according to the first question and/or determining context data corresponding to the first question according to the first question, and generating the reply content according to the reply style and the first question, wherein the method comprises generating the reply content according to the topic vector and/or the context data and the reply style and the first question.

In one embodiment of the disclosure, determining the topic vector corresponding to the first question according to the first question comprises determining the topic vector corresponding to the first question according to the first question in a historical interaction database, wherein the historical interaction database is constructed based on historical questions, historical replies corresponding to the historical questions, and identities of manual agents processing the historical questions.

In one embodiment of the disclosure, the obtaining the reply style of the target agent based on the first question and the identity of the target agent includes obtaining, in the historical interaction database, a first historical reply based on the first question and the identity of the target agent, wherein the first historical reply corresponds to the first question and the identity of the target agent, and/or corresponds to a second question and the identity of the target agent, the second question is a similarity question corresponding to the first question, the semantic similarity between the similarity question and the first question is higher than a preset semantic similarity threshold, and determining the reply style of the target agent according to the first historical reply.

In one embodiment of the disclosure, determining the reply style of the target agent based on the first historical reply includes performing word segmentation processing on the first historical reply to obtain a plurality of word segmentation results, determining a part-of-speech sequence of the first historical reply according to parts of speech of the plurality of word segmentation results, and obtaining the reply style of the target agent based on the part-of-speech sequence.

In one embodiment of the disclosure, the determining, in a historical interaction database, a topic vector corresponding to the first question according to the first question includes obtaining, in the historical interaction database, a second historical reply according to the first question, wherein the second historical reply corresponds to the first question and/or corresponds to the second question.

In one embodiment of the disclosure, before determining the topic vector corresponding to the first question based on the second historical answer, the method further comprises the steps of obtaining the length of the second historical answer, filtering the second historical answer according to a preset answer length range and the length of the second historical answer, wherein the preset answer length range is determined according to the average value and the variance of the length of the second historical answer, and determining the topic vector corresponding to the first question based on the second historical answer, wherein the method comprises the steps of determining the topic vector corresponding to the first question based on the filtered second historical answer.

In one embodiment of the disclosure, the determining the topic vector corresponding to the first question based on the second historical reply includes performing topic extraction on the second historical reply to obtain the topic vector corresponding to the first question.

In one embodiment of the disclosure, the extracting the topics from the second historical replies to obtain topic vectors corresponding to the first questions includes obtaining a first preset number of topics from the second historical replies, wherein each topic carries a corresponding weight, each topic includes a second preset number of phrases, obtaining a topic with the largest weight from the first preset number of topics according to the weight corresponding to each topic, and determining topic vectors corresponding to the first questions based on the second preset number of phrases included in the topic with the largest weight.

In one embodiment of the disclosure, the determining the topic vector corresponding to the first question based on the second preset number of phrases included in the topic with the largest weight includes performing vectorization processing on the second preset number of phrases included in the topic with the largest weight to obtain the second preset number of phrase vectors, accumulating the second preset number of phrase vectors to calculate an average value, and taking the average value as the topic vector corresponding to the first question.

In one embodiment of the disclosure, after determining the context data corresponding to the first question according to the first question, the method further comprises splicing the context data according to the sequence of the context data, deleting the spliced context data if the word number of the spliced context data is greater than the third preset number so that the word number of the deleted context data is less than or equal to the third preset number, and generating reply content according to the reply style and the first question, wherein the reply content is generated according to the theme vector and/or the deleted context data, and the reply style and the first question.

In one embodiment of the disclosure, generating reply content according to the topic vector and/or the context data, the reply style and the first question includes vectorizing the context data, the reply style and the first question to obtain a context data vector, a reply style vector and a first question vector, respectively, fusing the topic vector and/or the context data vector, the reply style vector and the first question vector, and generating the reply content based on the fused vectors.

In one embodiment of the disclosure, the fusing the topic vector and/or the context data vector, and the answer style vector and the first question vector includes stitching together the topic vector and/or the context data vector, and the answer style vector and the first question vector to obtain the fused vector.

In one embodiment of the disclosure, before generating the reply content according to the topic vector and/or the context data, the reply style and the first question, the method further comprises obtaining the reply content of the target agent to the first question, generating the reply content according to the topic vector and/or the context data, the reply style and the first question, and generating the reply content according to the reply content, the topic vector and/or the context data, the reply style and the first question.

In a second aspect of the embodiments of the present disclosure, there is provided a computer-readable storage medium having stored therein computer-executable instructions which, when executed by a processor, implement the session answer method as provided in the first aspect.

In a fourth aspect of the embodiment of the present disclosure, a model training apparatus is provided, which includes an obtaining module configured to obtain a first question, a first determining module configured to determine a target agent according to the first question, an obtaining module configured to obtain a reply style of the target agent based on the first question and an identity of the target agent, and a generating module configured to generate reply content according to the reply style and the first question.

In one embodiment of the disclosure, the method further comprises a second determining module, wherein the second determining module is used for determining a topic vector corresponding to the first question according to the first question before the generating module generates the reply content according to the reply style and the first question, and the generating module is specifically used for generating the reply content according to the topic vector and/or the context data and the reply style and the first question.

In one embodiment of the disclosure, the second determining module is specifically configured to determine, in a historical interaction database, a topic vector corresponding to the first question according to the first question, and/or determine, in the historical interaction database, context data corresponding to the first question according to the first question, where the historical interaction database is constructed based on a historical question, a historical reply corresponding to the historical question, and an identity of an artificial agent that handles the historical question.

In one embodiment of the disclosure, the obtaining module is specifically configured to obtain, in the historical interaction database, a first historical reply based on the first question and the identity of the target agent, where the first historical reply corresponds to the first question and the identity of the target agent, and/or corresponds to a second question and the identity of the target agent, where the second question is a similarity question corresponding to the first question, and a semantic similarity between the similarity question and the first question is higher than a preset semantic similarity threshold, and determine a reply style of the target agent according to the first historical reply.

In one embodiment of the disclosure, the obtaining module is specifically configured to perform word segmentation processing on the first historical reply to obtain a plurality of word segmentation results, determine a part-of-speech sequence of the first historical reply according to parts-of-speech of the plurality of word segmentation results, and obtain a reply style of the target agent based on the part-of-speech sequence.

In one embodiment of the disclosure, the second determining module is specifically configured to obtain, in the historical interaction database, a second historical reply according to the first question, where the second historical reply corresponds to the first question and/or corresponds to the second question, and determine a topic vector corresponding to the first question based on the second historical reply.

In one embodiment of the disclosure, the second determining module is specifically configured to obtain a length of the second historical reply, filter the second historical reply according to a preset reply length range and a length of the second historical reply, where the preset reply length range is determined according to an average value and a variance of the length of the second historical reply, and determine a topic vector corresponding to the first question based on the filtered second historical reply.

In one embodiment of the disclosure, the second determining module is specifically configured to perform topic extraction on the second historical reply to obtain a topic vector corresponding to the first question.

In one embodiment of the disclosure, the second determining module is specifically configured to obtain a first preset number of topics from the second historical replies, where each topic carries a corresponding weight, each topic includes a second preset number of phrases, obtain, according to the weight corresponding to each topic, a topic with a largest weight from the first preset number of topics, and determine a topic vector corresponding to the first question based on the second preset number of phrases included in the topic with the largest weight.

In one embodiment of the disclosure, the second determining module is specifically configured to perform vectorization processing on a second preset number of phrases included in the topic with the largest weight to obtain a second preset number of phrase vectors, accumulate the second preset number of phrase vectors, calculate an average value, and use the average value as a topic vector corresponding to the first problem.

In one embodiment of the disclosure, the second determining module is further configured to splice the context data according to the sequence of the context data, prune the spliced context data if the word count of the spliced context data is greater than the third preset number so that the word count of the pruned context data is less than or equal to the third preset number, and the generating module is specifically configured to generate reply content according to the topic vector and/or the pruned context data, the reply style and the first question.

In one embodiment of the disclosure, the generating module is specifically configured to perform vectorization processing on the context data, the reply style, and the first question to obtain a context data vector, a reply style vector, and a first question vector, fuse the topic vector and/or the context data vector, and the reply style vector and the first question vector, and generate the reply content based on the fused vectors.

In one embodiment of the disclosure, the generating module is specifically configured to splice the topic vector and/or the context data vector, and the answer style vector and the first question vector together to obtain the fused vector.

In one embodiment of the disclosure, the generating module is specifically configured to obtain replied content of the first question by the target agent, and generate replied content according to the replied content, the topic vector and/or the context data, and the replying style and the first question.

In a fourth aspect of the disclosed embodiments, a computing device is provided that includes at least one processor and a memory, the memory storing computer-executable instructions, the at least one processor executing the memory-stored computer-executable instructions causing the at least one processor to perform the session answer method as provided in the first aspect.

In the embodiment of the disclosure, the problem is acquired, and then, based on the problem, the target agent is determined, and the answer style of the target agent is obtained according to the problem and the identity of the target agent, so that the answer content is generated according to the answer style and the problem, the generated answer is consistent with the language style of the agent, the user experience is improved, the service quality is improved, the time for the artificial agent to modify the answer content to answer is reduced, and the service efficiency of the artificial agent is improved.

Drawings

The above, as well as additional purposes, features, and advantages of exemplary embodiments of the present disclosure will become readily apparent from the following detailed description when read in conjunction with the accompanying drawings. Several embodiments of the present disclosure are illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings, in which:

Fig. 1 schematically illustrates an application scenario schematic provided according to an embodiment of the present disclosure;

fig. 2 schematically illustrates a flow diagram of a session answer method provided according to an embodiment of the disclosure;

Fig. 3 schematically illustrates a flow diagram of a session answer method provided according to another embodiment of the disclosure;

fig. 4 schematically illustrates a flow diagram of a session answer method provided according to a further embodiment of the disclosure;

fig. 5 schematically illustrates a flow diagram of a session answer method provided according to a further embodiment of the disclosure;

FIG. 6 schematically illustrates reply content generation schematic provided in accordance with an embodiment of the present disclosure;

FIG. 7 schematically illustrates a structural diagram of a storage medium provided according to an embodiment of the present disclosure;

Fig. 8 schematically illustrates a structural diagram of a session answering device provided according to an embodiment of the present disclosure;

fig. 9 schematically illustrates a structural diagram of a session answering device provided according to another embodiment of the present disclosure;

fig. 10 schematically illustrates a structural schematic diagram of a computing device provided according to an embodiment of the present disclosure.

In the drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

Detailed Description

The principles and spirit of the present disclosure will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are presented merely to enable one skilled in the art to better understand and practice the present disclosure and are not intended to limit the scope of the present disclosure in any way. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

Those skilled in the art will appreciate that embodiments of the present disclosure may be implemented as a system, apparatus, device, method, or computer program product. Accordingly, the present disclosure may be embodied in the form of entirely hardware, entirely software (including firmware, resident software, micro-code, etc.) or in a combination of hardware and software.

According to the embodiment of the disclosure, a session response method, a session response medium, a session response device and a session response computing device are provided.

Furthermore, any number of elements in the figures is for illustration and not limitation, and any naming is used for distinction only and not for any limiting sense.

The principles and spirit of the present disclosure are explained in detail below with reference to several representative embodiments thereof.

Summary of The Invention

The inventor finds that with the continuous development of society, the demands of users are more and more diversified, and the problems are more and more increased in the process of interacting with intelligent clients, but intelligent customer service can only be used for replying a few simple repeated problems, and aiming at the phenomena of large matching error and low reply correctness when the problems are more complicated, the user needs to transfer a manual agent to reply correspondingly. The relative shortage of the manual seat staff needs to be combined with a customer service system, for example, the customer service system generates a reference answer based on the questions of the user, and the manual seat refers to the answer to answer the questions of the user. However, when the manual agent refers to the answer generated by the customer service system and answers the questions presented by the user, a language style change often occurs, for example, the user feels that the agent replying to the questions before and the agent replying to the questions after are not the same person, so that the user experience is poor and the service quality is affected.

In order to improve the thought of the service quality of the artificial agent, in the embodiment of the disclosure, the answer style of the agent is considered, so that the answer generated by the client system is consistent with the language style of the agent, the user experience is improved, and the service quality is improved.

Having described the basic principles of the present disclosure, various non-limiting embodiments of the present disclosure are specifically described below.

Application scene overview

The scenes to which the embodiments of the present disclosure are applicable include a session answer scene.

Referring first to fig. 1, fig. 1 schematically shows an application scenario schematic diagram provided according to an embodiment of the present disclosure, where an apparatus involved in the application scenario includes a first terminal 101, a server 102, and a second terminal 103. The first terminal 101 and the server 102 may communicate via a network, and the second terminal 103 and the server 102 may also communicate via a network. At this time, the first terminal 101 may be a terminal of the user, and the second terminal 103 may be a terminal of the target agent. In addition, the first terminal 101 and the second terminal 103 may be deployed with applications related to session replies.

When the application scenario is a session answer, the user may open an application program related to the session answer on the first terminal 101, and interact with the application program, for example, input a problem that an agent answer is required. The server 102 obtains the question, determines a target agent according to the question, further obtains a reply style of the target agent based on the question and an identity of the target agent, generates reply content according to the reply style and the question so that the generated reply is consistent with a language style of the target agent, and then transmits the reply content to the second terminal 103. Furthermore, the target agent can look up the reply content at the second terminal 103, open the application program related to the session reply, reply the question based on the reply content, and the user can look up the reply of the target agent at the first terminal 101, so that the user experience is better, and the service quality of the agent is improved.

Exemplary method

A session response method provided according to an exemplary embodiment of the present disclosure is described below with reference to fig. 2 to 6 in conjunction with the application scenario of fig. 1. It should be noted that the above application scenario is only shown for the convenience of understanding the spirit and principles of the present disclosure, and the embodiments of the present disclosure are not limited in any way in this respect. Rather, embodiments of the present disclosure may be applied to any scenario where applicable.

It should be noted that, the embodiments of the present disclosure may be applied to an electronic device, and the electronic device may be a terminal or a server, that is, the session response method provided by the exemplary embodiments of the present disclosure may be executed on the terminal or the server. For example, the electronic device is a server, and after the reply content is generated, the reply content may be sent to the terminal of the target agent, so that the target agent quickly replies to the problem based on the reply content. And after the reply content is generated, the target agent can check the reply content to reply, so that the condition of language style change is reduced, the user experience is improved, and the service quality is improved.

The terminal may be a Personal Digital Assistant (PDA) device, a handheld device with a wireless communication function (e.g., a smart phone, a tablet computer), a computing device (e.g., a personal computer (personal computer, PC)), an in-vehicle device, a wearable device (e.g., a smart watch, a smart bracelet), a smart home device (e.g., a smart display device), etc.

The servers may be monolithic servers or distributed servers across multiple computers or computer data centers. The server may also be of various types, such as, but not limited to, a web server, an application server, or a database server, or a proxy server.

Alternatively, the server may comprise hardware, software, or embedded logic components or a combination of two or more such components for performing the appropriate functions supported or implemented by the server. For example, a server, such as a blade server, cloud server, etc., or may be a server group consisting of multiple servers, may include one or more of the above-mentioned classes of servers, etc.

It should be noted that, according to the session response method provided in the exemplary embodiment of the present disclosure, the session response method may be performed on the same device or may be performed on a different device.

Referring to fig. 2, fig. 2 schematically shows a flow diagram of a session answer method provided according to an embodiment of the disclosure. The execution body of the embodiment may be determined according to an actual application scenario, which is not particularly limited in the embodiment of the present disclosure. As shown in fig. 2, the session response method includes:

S201, acquiring a first problem.

The first problem may be determined according to the actual situation, for example, referring to the following table 1, the content of the session between the visitor and the seat may be known, where the first problem is a statement corresponding to the sequence number 6, that is, "no point for use by another watch is" yes ".

TABLE 1

S202, determining a target seat according to the first problem.

Here, after the first problem is acquired, the present embodiment may determine the target seat according to the first problem. As described in table 1, the problem of the number 6 is acquired, and the problem is regarded as the first problem, and further, the agent interacting with the visitor in table 1 is regarded as the target agent.

S203, obtaining the reply style of the target agent based on the first question and the identity of the target agent.

In this embodiment, in order to accurately determine the reply style of the target agent, a first historical reply may be obtained in the historical interaction database based on the first question and the identity of the target agent, and then the reply style of the target agent may be determined according to the first historical reply.

The history interaction database is constructed based on the history questions, the history answers corresponding to the history questions, and the identities of the manual agents for processing the history questions.

For example, when the above-mentioned history interaction database is constructed, the history questions, the history answers corresponding to the history questions, and the identities of the manual agents handling the history questions may be acquired, and the above-mentioned history interaction database may be constructed based on the acquired history questions, the history answers corresponding to the history questions, and the identities of the manual agents handling the history questions.

Here, the above-mentioned history questions, the history replies corresponding to the history questions, and the identities of the artificial agents that process the history questions may be determined by collecting the conversation corpus of all the artificial agents that have ended within a preset period of time. For example, from the collected conversation corpus, continuous sentences of the same role (visitor or agent) are combined, and conversation with visitor and agent at intervals is obtained after the combination.

For example, the visitor continues to say "manual", "my 1000 points cannot be held against money", and then incorporate it as "manual", "my 1000 points cannot be held against money". After merging, the merged expression of the visitor is used as a history question, and the answer of the visitor expression back seat is used as a history answer corresponding to the history question, for example, "you can check whether there are goods liked by you on the point page, and can use a point trading purchase mode" as the history answer corresponding to the history question. The conversation corpus of the artificial seat A and a user is taken as an example, and the conversation corpus comprises the identity of the artificial seat A, such as the number, the work number and the like of the artificial seat A. The embodiment can determine the history questions and the history answers corresponding to the history questions from the collected conversation corpus, and then determine the identity of the artificial seat for processing the history questions according to the identity of the artificial seat included in the conversation corpus.

In order to accurately determine the historical questions and the historical answers corresponding to the historical questions, after the conversation corpus is collected, the conversation corpus can be filtered, for example, incomplete conversation corpus is filtered (for example, conversation processed by other people is transferred), some images, videos, voice contents and the like in the conversation corpus are filtered, and the filtered images, videos, voice contents and the like can be set according to actual conditions (for example, expression images, videos, voice contents and the like without practical significance), so that the filtered conversation corpus is complete and accurate, and the accuracy of subsequent processing results is improved.

In addition, in this embodiment, the first history answer may correspond to the identity of the first question and the target agent, and/or correspond to the identity of the second question and the identity of the target agent, where the second question is a similarity question corresponding to the first question, and the similarity between the similarity question and the first question is higher than a preset semantic similarity threshold. The first problem is that "the other wristwatch cannot be used for the point is the same as the first problem", the second problem is similar to the first problem, and the first problem may be "which watches can be used for the point is also used for the point", or "which watches can be used for the point are used for the point".

Here, the first history answer may be a history answer corresponding to the first question and the identity of the target agent in the history interaction database, and/or a history answer corresponding to the second question and the identity of the target agent in the history interaction database, the second question being a similar question corresponding to the first question. In this way, the number of the first historical replies acquired from the historical interaction database is more, the content is richer, and the reply style of the target agent can be completely and accurately determined based on the first historical replies. The semantic similarity threshold may be set according to practical situations, for example, 60%. If the voice similarity between the second question and the first question is higher than a preset semantic similarity threshold, such as 60%, the second question is a similarity question corresponding to the first question.

Alternatively, when determining the response style of the target agent based on the first historical response, word segmentation may be performed on the first historical response to obtain a plurality of word segmentation results, and then a part-of-speech sequence of the first historical response may be determined according to parts of speech of the plurality of word segmentation results, so that the response style of the target agent is obtained based on the part-of-speech sequence.

If the first question is "no credit available to other watches is yes", the first history answer obtained is "credit support bridge". The embodiment performs word segmentation processing on the target seat to obtain a plurality of word segmentation results, such as 'integral', 'support', 'deduction', 'couple', and the corresponding parts of speech are 'noun', 'verb', 'imaginary word', and further determines the part of speech sequence of the first historical reply according to the parts of speech of the plurality of word segmentation results, thereby obtaining the reply style of the target seat.

S204, generating reply content according to the reply style and the first question.

Here, the answer style and the first question may be vectorized to obtain an answer style vector and a first question vector, and the answer style vector and the first question vector may be fused to generate the answer content based on the fused vectors. The foregoing reply style and first question may be vectorized, for example, using a predetermined encoder (e.g., convolutional neural network (Convolutional Neural Networks, CNN), cyclic neural network (Recurrent Neural Network, RNN), transducer, etc.).

For example, the embodiment may splice the answer style vector and the first question vector together to obtain the fused vector.

Optionally, after obtaining the fused vector, the fused vector may be input to a preset decoder (such as RNN, long Short-Term Memory (LSTM), transform, etc.) to generate the reply content, where the decoder is configured to generate text content based on the vector.

In addition, before the reply content is generated, the embodiment may further obtain the reply content of the target agent to the first question, for example, "support" the content in sequence number 8 in table 1, and further generate complete and accurate reply content according to the reply content, the reply style and the first question. The replied content may be vectorized to obtain a replied content vector, the replying style vector and the first question vector are fused, and reply content is generated based on the fused vector, e.g., the replied content vector, the replying style vector and the first question vector are spliced together to obtain the fused vector, so as to generate the reply content.

According to the method and the device for processing the response of the target agent, the target agent is determined based on the problem, and the response style of the target agent is obtained according to the problem and the identity of the target agent, so that the response content consistent with the language style of the agent is generated according to the response style and the problem, the user experience is good, the service quality is improved, the time for the artificial agent to modify the response content to respond is reduced, and the service efficiency of the artificial agent is improved.

In addition, when the existing manual agents refer to the answers generated by the customer service system and answer questions presented by users, answer errors often occur, and the accuracy is low. Therefore, in order to reduce the situation of response errors and improve response accuracy, before generating response content according to the response style and the first question, the embodiment of the disclosure further considers determining a topic vector corresponding to the first question according to the first question and/or determining context data corresponding to the first question according to the first question, thereby generating complete and accurate response content according to the topic vector and/or the context data and the response style and the first question, improving the response generated by a manual seat reference customer service system, improving the accuracy of responding to the questions of a user, enabling the generated response content to be consistent with the language style of the seat, improving user experience and improving service quality.

Fig. 3 is a flowchart of a session answering method according to another embodiment of the present disclosure, as shown in fig. 3, in this embodiment, before generating answer content according to the answer style and the first question, determining a topic vector corresponding to the first question according to the first question, so as to generate answer content according to the topic vector, the answer style and the first question, where the method includes:

S301, acquiring a first problem.

S302, determining a target seat according to the first problem.

S303, obtaining the reply style of the target agent based on the first question and the identity of the target agent.

The implementation manner of steps S301 to S303 is referred to the related description in the embodiment of fig. 2, and will not be repeated here.

S304, according to the first problem, determining a theme vector corresponding to the first problem.

Here, in the history interaction database, a topic vector corresponding to the first question may be determined according to the first question.

For example, the present embodiment may obtain, in the history interaction database, a second history answer according to the first question, where the second history answer corresponds to the first question and/or corresponds to the second question, so as to determine, based on the second history answer, a topic vector corresponding to the first question. The second question is a similarity question corresponding to the first question, and the semantic similarity between the similarity question and the first question is higher than the preset semantic similarity threshold.

Here, in the history interaction database, answers of all agents corresponding to the first question and/or corresponding to the second question are acquired as the second history answer, so that the content of the acquired second history answer is more comprehensive, and thus, a topic vector corresponding to the first question can be accurately determined based on the acquired second history answer later.

Optionally, before determining the topic vector corresponding to the first question based on the second historical reply, the present embodiment may further obtain a length of the second historical reply, and further filter the second historical reply according to a preset reply length range and the length of the second historical reply, so as to determine the topic vector corresponding to the first question based on the filtered second historical reply.

Wherein the preset reply length range is determined according to the average value and the variance of the lengths of the second historical replies. For example, by counting the length of the second history answer, which can be understood as how many words are included in the second history answer, and further, calculating the average μ and variance σ of the lengths, and determining the shortest length asMaximum length ofI.e. determining how many words are least included and how many words are most included in said second historical reply, thereby determining said preset reply length range based on said shortest length and maximum length. For example, if the server performs statistics, and the average value of the lengths of the second historical replies is 5.4 and the variance is 1.1, the shortest length is determined to be(Rounded upward), maximum length of (Round down) it is further determined that the preset reply length range is 4-7. The first question is "no credit available for other watches" and the second history is "no clear available for other watches" length 12. When the second historical replies are filtered, the preset reply length range may be determined based on the shortest length and the maximum length, and further, the second historical replies may be filtered according to the preset reply length range, where the shortest length is 4, the maximum length is 7, the preset reply length range is 4-7, the length of the second historical replies is 12, and the second historical replies may be deleted in the embodiment not within the range of 4-7.

Here, the present embodiment considers that the quality of the answers with too short or too long length is often not high, so that a range of answer lengths is determined according to the average value and the variance of the lengths of the second historical answers, answers with answer lengths not within the range of answer lengths are all deleted, the answer quality is improved, and further, a topic vector corresponding to the first question is determined based on the filtered second historical answer, and the quality of the topic vector is improved.

In this embodiment, when determining the topic vector corresponding to the first question based on the second historical reply, topic extraction may be performed on the second historical reply to obtain the topic vector corresponding to the first question. In an exemplary embodiment, a first preset number of topics may be obtained from the second historical replies, where each topic carries a corresponding weight, each topic includes a second preset number of phrases, and further, according to the weight corresponding to each topic, a topic with the largest weight is obtained from the first preset number of topics, so that a topic vector corresponding to the first problem is determined based on the second preset number of phrases included in the topic with the largest weight. Here, the first preset number and the second preset number may be determined according to practical situations, for example, the first preset number is 3, and the second preset number is also 3.

Alternatively, the second preset number of phrases included in the topic with the largest weight may be vectorized to obtain the second preset number of phrase vectors, the second preset number of phrase vectors are accumulated to calculate an average value, and the average value is used as the topic vector corresponding to the first problem.

In the case of extracting the subject from the history reply, the following description will be given by taking the following sentence as an example of extracting the subject:

Automobile A is a world-known luxury automobile brand;

automobile B, national automobile brand;

And the automobile C, the company headquarters are in the city of Yi province, and the business spans four industries of automobiles, rail transit, new energy and electronics.

In this embodiment, before extracting the topics from the sentences, a first preset number of topics may be extracted, and each extracted topic may include a second preset number of phrases, where the first preset number is 3, and the second preset number is also 3. When extracting the topics from the sentence, firstly determining a plurality of topics contained in the sentence, then determining the probability distribution of each word in the sentence belonging to each topic in the plurality of topics and the probability distribution of each word on each topic, further determining the weight of each topic based on the probability distribution of each word in the sentence belonging to each topic in the plurality of topics, sorting from large to small according to the weight of each topic, acquiring the first 3 topics from the sorting result, acquiring the topics 1, 2 and 3 as shown in the following table 2, the weight of the topic 1 is 0.45, the weight of the topic 2 is 0.35, the weight of the topic 3 is 0.2, and the words on each topic are ranked based on the probability distribution of the words on each topic, for example, the words on each topic are ranked according to the probability distribution of the words on each topic from big to small, the first 3 words are obtained in the ranking result, as shown in the following table 2, the first 3 words in the topic 1 are obtained as automobile a, automobile B and automobile C, the first 3 words in the topic 2 are obtained as first country, ethyl province and propyl city, and the first 3 words in the topic 3 are obtained as automobiles, electrons and new energy sources.

TABLE 2

Theme 1 (0.45)	Theme 2 (0.35)	Theme 3 (0.2)
			Automobile A	First country	Automobile
Automobile B	Easter B province	Electronic device
			Automobile C	Prop City of third	New energy source

Alternatively, in this embodiment, the topic with the largest weight, that is, topic 1, may be obtained from the 3 topics according to the weight corresponding to each topic, so that, based on 3 phrases included in topic 1, a topic vector corresponding to the first problem is determined, for example, 3 phrases (automobile a, automobile B, and automobile C) included in topic 1 are subjected to vectorization processing, for example, 3 phrases included in topic 1 are respectively input into a vectorization model, and the vectorization model is used for converting text into a vector expressing text semantics, and further, 3 phrase vectors corresponding to 3 phrases included in topic 1 are obtained. The vector dimension obtained by the conversion of the vectorization model may be determined according to the actual situation, where, taking the example that the vector dimension obtained by the conversion of the vectorization model is a four-dimensional vector, a phrase vector corresponding to the phrase car a is [0.0115525,0.71636848,0.75681735,0.61298513], a phrase vector corresponding to the phrase car B is [0.29095593,0.79231239,0.08199059,0.56924747], and a phrase vector corresponding to the phrase car C is [0.26886547,0.41897932,0.25461575,0.0446018]. Further, the present embodiment may calculate an average value ([ 0.19045797,0.6425534,0.36447456,0.4089448 ]) after accumulating the above 3 phrase vectors, and use the average value as the topic vector corresponding to the above first problem.

Here, the embodiment obtains the topic vector corresponding to the first question, so that a more comprehensive reply content can be generated later according to the topic vector, the reply style and the first question, so that the manual agent can completely and accurately reply to the question raised by the user based on the reply content, the service quality is improved, the workload of the agent is also reduced, and the cost is reduced.

And S305, generating reply content according to the theme vector, the reply style and the first question.

Here, the answer style and the first question may be vectorized to obtain an answer style vector and a first question vector, respectively, and the subject vector, the answer style vector and the first question vector may be fused, and the answer content may be generated based on the fused vectors. For example, the server may splice the topic vector, the answer style vector, and the first question vector together to obtain the fused vector.

In addition, before the reply content is generated, the embodiment may further obtain the reply content of the target agent to the first question, and further generate complete and accurate reply content according to the reply content, the topic vector, the reply style and the first question. The server may perform vectorization processing on the replied content to obtain a replied content vector, fuse the replied content vector, the topic vector, the replying style vector, and the first question vector, and generate replied content based on the fused vector. And the server splices the replied content vector, the theme vector, the replying style vector and the first question vector together to obtain a fused vector, and generates replied content.

In the embodiment of the disclosure, before generating the reply content according to the reply style and the first question, determining the topic vector corresponding to the first question according to the first question is also considered, so that according to the topic vector, the reply style and the first question, more comprehensive reply content is generated, a manual agent can completely and accurately reply to the question raised by the user based on the reply content, the reply accuracy is improved, the generated reply content is consistent with the language style of the agent, the user experience is improved, and the service quality is improved.

In addition, fig. 4 is a flowchart of a session response method according to still another embodiment of the present disclosure, as shown in fig. 4, in this embodiment, before generating a response content according to the response style and the first question, context data corresponding to the first question is determined according to the first question, so that the response content is generated according to the context data, the response style and the first question, and the method includes:

S401, acquiring a first problem.

S402, determining a target seat according to the first problem.

S403, obtaining the reply style of the target agent based on the first question and the identity of the target agent.

The implementation manner of steps S401 to S403 is referred to the related description in the embodiment of fig. 2, and will not be repeated here.

S404, according to the first problem, determining context data corresponding to the first problem.

Here, after the above-described first problem is obtained, corresponding context data may be determined according to the above-described first problem. As described above, in table 1, the problem of the number 6 is obtained, and the problem is taken as the first problem, and further, the session in which the numbers 1 to 5 in table 1 have occurred is taken as the context data corresponding to the first problem, and the session in which the number 7 in table 1 has occurred is taken as the context data corresponding to the first problem, so that the context data corresponding to the first problem is determined.

Optionally, after determining the context data corresponding to the first problem, the present embodiment may further splice the context data according to the sequence of the context data, and if the number of words of the spliced context data is greater than a third preset number, prune the spliced context data so that the number of words of the pruned context data is less than or equal to the third preset number.

Taking the session with the sequence numbers 1-5 in the table 1 as the above data corresponding to the first problem, the session with the sequence number 7 in the table 1 as the above data corresponding to the first problem, for example, the server splices the sessions with the sequence numbers 1-5 and 7 according to the sequence of the sessions with the sequence numbers 1-5 and 7 in the table 1 to obtain a long sentence, namely "manual, my 1000 points cannot be used for money, you can check whether there is a favorite commodity in the point page, the point exchange purchase mode can be used instead of 100 yuan for one minute, and meanwhile, if other watches can be used for points, how to use the watches. Further, if the word number of the long sentence after splicing is greater than the third preset number, deleting the long sentence after splicing. Here, the third preset number may be determined according to actual situations, for example, 128 words.

If the number of words of the spliced long sentence is greater than a third preset number, for example, greater than 128 words, it is indicated that the content of the spliced long sentence is more, and if the spliced long sentence is directly used for subsequent processing, the processing speed may be slower, so that the spliced long sentence is pruned, so that the number of words of the pruned long sentence is less than or equal to the third preset number, for example, less than or equal to 128 words. Optionally, in this embodiment, when deleting the spliced long sentence, deleting the spliced long sentence according to a preset deletion word, so that the number of words of the deleted long sentence is smaller than or equal to the third preset number, where the preset deletion word may be set according to an actual situation, such as a word of a mood having no practical meaning. In addition, when deleting the spliced long sentence, deleting the spliced long sentence according to a preset position, so that the number of words of the deleted long sentence is smaller than or equal to the third preset number, where the preset position may be set according to an actual situation, for example, a starting position of the spliced long sentence is set as the preset position.

After determining the context data corresponding to the first problem, the embodiment may further splice, prune, etc. the context data, so as to improve the subsequent processing speed, thereby improving the service efficiency of the manual seat.

S405, generating reply content according to the context data, the reply style and the first question.

Here, the server may perform vectorization processing on the context data, the reply style, and the first question, respectively, to obtain a context data vector, a reply style vector, and a first question vector, and fuse the context data vector, the reply style vector, and the first question vector, and generate the reply content based on the fused vectors. For example, the server may splice the context data vector, the reply style vector, and the first question vector together to obtain the fused vector. In the vectorization process, the vectorization process is described with respect to the context data, and the present embodiment may vectorize the context data using a vectorization model. Here, the vectorization model is used to convert text into a vector expressing text semantics, and when vectorizing the context data by using the vectorization model, the context data may be input into the vectorization model to obtain a context data vector corresponding to the context data. The vector dimension obtained by converting the vectorization model may be determined according to the actual situation, where the vector dimension obtained by converting the vectorization model is a two-dimensional vector, for example, the context data is "manual, my 1000 points cannot support money, you can check whether there are favorite goods on the point page, the point exchange purchase mode may be used instead of 100-element one-minute mode, and meanwhile, if another watch may use the point, input the two-dimensional vector into the vectorization model, and obtain the context data vector corresponding to the context data as [0.48539348,0.34782834].

In addition, before the reply content is generated, the embodiment may further obtain the reply content of the target agent to the first question, and further generate complete and accurate reply content according to the reply content, the context data, the reply style and the first question. The replied content may be vectorized to obtain a replied content vector, the context data vector, the replied style vector, and the first question vector may be fused, and the replied content may be generated based on the fused vector. And if the replied content vector, the context data vector, the replying style vector and the first question vector are spliced together, a fused vector is obtained, and replying content is generated.

In the embodiment of the disclosure, before generating the reply content according to the reply style and the first question, determining the context data corresponding to the first question according to the first question is also considered, so that more complete reply content is generated according to the context data, the reply style and the first question, the accuracy of a manual agent for replying to the user question based on the reply content is improved, and the generated reply is consistent with the language style of the agent, so that the user experience is better, and the service quality is improved.

Fig. 5 is a flowchart of a session answering method according to another embodiment of the present disclosure, as shown in fig. 5, in this embodiment, before generating answer content according to the answer style and the first question, determining a topic vector corresponding to the first question according to the first question, and determining context data corresponding to the first question according to the first question, so as to generate answer content according to the topic vector, the context data, and the answer style and the first question, where the method includes:

S501, acquiring a first problem.

S502, determining a target seat according to the first problem.

The implementation manner of steps S501-S502 is described with reference to the embodiment of fig. 2, and is not described herein.

S503, obtaining the reply style of the target agent based on the first question and the identity of the target agent.

In this embodiment, a first historical answer may be obtained in the historical interaction database based on the first question and the identity of the target agent, and then, according to the first historical answer, the answer style of the target agent may be determined.

S504, according to the first problem, determining a theme vector corresponding to the first problem, and according to the first problem, determining context data corresponding to the first problem.

In addition, after determining the context data corresponding to the first problem, the present embodiment may splice the context data according to the order of the context data, and prune the context data after splicing. If the word number of the spliced context data is greater than a third preset number, the server prunes the spliced context data so that the word number of the pruned context data is less than or equal to the third preset number.

S505, generating reply content according to the topic vector, the context data, the reply style and the first question.

For example, as shown in fig. 6, the present embodiment may first construct a history interaction database, obtain a history question, a history answer corresponding to the history question, and an identity of a manual agent that handles the history question, and construct the history interaction database based on the obtained history question, the history answer corresponding to the history question, and the identity of the manual agent that handles the history question.

Then, a first historical answer is obtained in the historical interaction database based on the first question and the identity of the target agent, and then the answer style of the target agent is determined according to the first historical answer.

When determining the reply style of the target agent based on the first historical reply, word segmentation processing may be performed on the first historical reply to obtain a plurality of word segmentation results, and then a part-of-speech sequence of the first historical reply may be determined according to the parts of speech of the plurality of word segmentation results, so that the reply style of the target agent is obtained based on the part-of-speech sequence. The first question is "no credit available to other watches is yes", and the first history is obtained as "credit support log". The embodiment performs word segmentation processing on the target seat to obtain a plurality of word segmentation results, such as 'integral', 'support', 'deduction', 'couple', and the corresponding parts of speech are 'noun', 'verb', 'imaginary word', and further determines the part of speech sequence of the first historical reply according to the parts of speech of the plurality of word segmentation results, thereby obtaining the reply style of the target seat.

Optionally, in the history interaction database, a second history answer may be obtained according to the first question, and the second history answer may be filtered, for example, a length of the second history answer is obtained, and further, the second history answer is filtered according to a preset answer length range and a length of the second history answer. After the filtering is completed, the server may perform topic extraction on the filtered second historical reply, to obtain a topic vector corresponding to the first question.

In an exemplary embodiment, a first preset number of topics may be obtained from the second historical replies, where each topic carries a corresponding weight, each topic includes a second preset number of phrases, and further, according to the weight corresponding to each topic, a topic with the largest weight is obtained from the first preset number of topics, so that a topic vector corresponding to the first problem is determined based on the second preset number of phrases included in the topic with the largest weight.

Optionally, after the first problem is obtained, corresponding context data may be determined according to the first problem, and the context data may be spliced according to the sequence of the context data, and if the number of words of the spliced context data is greater than a third preset number, the spliced context data is pruned, so that the number of words of the pruned context data is less than or equal to the third preset number.

Here, after the above-mentioned answer style, the topic vector, and the context data are obtained, the present embodiment may perform vectorization processing on the above-mentioned context data, answer style, and first question, respectively, to obtain a context data vector, an answer style vector, and a first question vector, and fuse the above-mentioned topic vector, context data vector, and answer style vector with the first question vector, and generate the above-mentioned answer content based on the fused vectors. For example, the topic vector, the context data vector, the answer style vector, and the first question vector may be stitched together to obtain the fused vector.

In the vectorization process, the present embodiment may use a vectorization model to vectorize the context data vector, the reply style vector, and the first question vector. Here, the vectorization model described above is used to convert text into vectors that express text semantics. In the vectorization process, the present embodiment may input the context data vector, the answer style vector, and the first question vector into the vectorization model to obtain a context data vector corresponding to the context data, an answer style vector corresponding to the answer style, and a first question vector corresponding to the first question. The vector dimension obtained by converting the vectorization model may be determined according to an actual situation, where, taking the vector dimension obtained by converting the vectorization model as a two-dimensional vector as an example, a context data vector corresponding to the context data is obtained as [0.48539348,0.34782834], a reply style vector corresponding to the reply style is obtained as [0.87664415,0.93101361], and a first question vector corresponding to the first question is obtained as [0.92049985,0.17425929]. If the topic vector is known to be [0.38434537,0.75824908], the present embodiment may splice the topic vector, the context data vector, and the answer style vector with the first question vector to obtain the fused vector [0.48539348,0.34782834,0.38434537,0.75824908,0.87664415,0.93101361,0.92049985,0.17425929].

Further, as shown in fig. 6, after the fused vector is obtained, the fused vector may be input to the preset decoder to generate the reply content.

In addition, before the reply content is generated, the reply content of the target agent to the first question may be obtained, and further, a more complete and accurate reply content may be generated according to the reply content, the topic vector, the context data, the reply style and the first question. The replied content may be vectorized to obtain a replied content vector, the topic vector, the context data vector, the replying style vector, and the first question vector may be fused, and the replied content may be generated based on the fused vector. And if the replied content vector, the theme vector, the context data vector, the replying style vector and the first question vector are spliced together, a fused vector is obtained, and replying content is generated.

Compared with the prior art, the method and the device for generating the answer content have the advantages that the answer style of the target agent is obtained, and according to the answer style and the questions, the answer content is generated, so that the generated answer is consistent with the language style of the agent, the user experience is improved, and the service quality is improved. In addition, before generating the reply content according to the reply style and the questions, the embodiment also considers determining the topic vector corresponding to the questions and the context data corresponding to the questions according to the questions, so that complete and accurate reply content is generated according to the topic vector and the context data and the reply style and the questions, the response generated by the manual agent reference customer service system is improved, and the accuracy of responding to the questions presented by the user is improved. In addition, before generating the reply content, the embodiment also considers the replied content of the target agent to the question, and further generates more complete and accurate reply content according to the replied content, the topic vector, the context data, the reply style and the question, thereby further improving the reply accuracy.

Exemplary Medium

Having described the method of the exemplary embodiments of the present disclosure, next, a storage medium of the exemplary embodiments of the present disclosure will be described with reference to fig. 7.

Referring to fig. 7, a storage medium 70, in which a program product for implementing the above-described method according to an embodiment of the present disclosure is stored, may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may be run on a terminal device such as a personal computer. However, the program product of the present disclosure is not limited thereto.

The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of a readable storage medium include an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The readable signal medium may include a data signal propagated in baseband or as part of a carrier wave with readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. The readable signal medium may also be any readable medium other than a readable storage medium.

Program code for carrying out operations of the present disclosure may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, partly on a remote computing device, or entirely on the remote computing device or server. In the context of remote computing devices, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN).

Exemplary apparatus

Having described the medium of the exemplary embodiment of the present disclosure, the session response device of the exemplary embodiment of the present disclosure is described with reference to fig. 8 to fig. 9, where the session response device is used to implement the session response method provided by any one of the method embodiments, and the implementation principle and technical effect are similar, and are not repeated herein.

Referring to fig. 8, fig. 8 is a schematic structural diagram of a session answering device according to an embodiment of the present disclosure. As shown in fig. 8, the session response device includes:

An obtaining module 801 is configured to obtain a first problem.

A first determining module 802, configured to determine a target agent according to the first problem.

And an obtaining module 803, configured to obtain a reply style of the target agent based on the first question and the identity of the target agent.

A generating module 804, configured to generate reply content according to the reply style and the first question.

Referring to fig. 9, fig. 9 is a schematic structural diagram of a session answering device according to another embodiment of the present disclosure. As shown in fig. 9, the session response device further includes:

A second determining module 805, configured to determine, before the generating module 804 generates reply content according to the reply style and the first question, a topic vector corresponding to the first question according to the first question, and/or determine, according to the first question, context data corresponding to the first question.

The generating module 804 is specifically configured to generate reply content according to the topic vector and/or the context data, and the reply style and the first question.

In one embodiment of the present disclosure, the second determining module 805 is specifically configured to:

And determining a topic vector corresponding to the first question in a historical interaction database according to the first question, wherein the historical interaction database is constructed based on the historical question, a historical reply corresponding to the historical question and the identity of a manual seat for processing the historical question.

In yet another embodiment of the present disclosure, the obtaining module 803 is specifically configured to:

And acquiring a first historical reply based on the first question and the identity of the target agent in the historical interaction database, wherein the first historical reply corresponds to the first question and the identity of the target agent and/or corresponds to a second question and the identity of the target agent, the second question is a similar question corresponding to the first question, the semantic similarity between the similar question and the first question is higher than a preset semantic similarity threshold value, and determining the reply style of the target agent according to the first historical reply.

the method comprises the steps of carrying out word segmentation on a first historical answer to obtain a plurality of word segmentation results, determining a part-of-speech sequence of the first historical answer according to the parts of speech of the plurality of word segmentation results, and obtaining the answer style of the target seat based on the part-of-speech sequence.

In yet another embodiment of the present disclosure, the second determining module 805 is specifically configured to:

acquiring a second historical reply according to the first question in the historical interaction database, wherein the second historical reply corresponds to the first question and/or corresponds to the second question; a topic vector corresponding to the first question is determined based on the second historical answer.

The method comprises the steps of obtaining the length of a first historical reply, obtaining the length of the first historical reply, filtering the first historical reply according to a preset reply length range and the length of the first historical reply, wherein the preset reply length range is determined according to the average value and the variance of the length of the first historical reply, and determining a topic vector corresponding to a first question based on the filtered first historical reply.

and extracting the subject from the second historical answer to obtain a subject vector corresponding to the first question.

The method comprises the steps of obtaining a first preset number of topics from the second historical answers, obtaining a topic with the largest weight from the first preset number of topics according to the weight corresponding to each topic, and determining a topic vector corresponding to the first problem based on the second preset number of phrases included in the topic with the largest weight.

the method comprises the steps of carrying out vectorization processing on a second preset number of phrases included in a theme with the largest weight to obtain phrase vectors with the second preset number, accumulating the phrase vectors with the second preset number, calculating an average value, and taking the average value as a theme vector corresponding to the first problem.

In yet another embodiment of the present disclosure, the second determining module 805 is further configured to:

and if the word number of the spliced context data is larger than a third preset number, deleting the spliced context data so that the word number of the deleted context data is smaller than or equal to the third preset number.

The generating module 804 is specifically configured to:

And generating reply content according to the topic vector and/or the deleted context data, the reply style and the first question.

In yet another embodiment of the disclosure, the generating module 804 is specifically configured to:

carrying out vectorization processing on the context data, the reply style and the first question respectively to obtain a context data vector, a reply style vector and a first question vector; and fusing the topic vector and/or the context data vector, the reply style vector and the first question vector, and generating the reply content based on the fused vector.

And splicing the theme vector and/or the context data vector, the reply style vector and the first question vector together to obtain the fused vector.

and generating reply content according to the replied content, the topic vector and/or the context data, the reply style and the first question.

Exemplary computing device

Having described the methods, media, and apparatus of exemplary embodiments of the present disclosure, a computing device of exemplary embodiments of the present disclosure is next described with reference to fig. 10.

The computing device 100 shown in fig. 10 is only one example and should not be taken as limiting the functionality and scope of use of embodiments of the present disclosure.

As shown in fig. 10, the computing device 100 is in the form of a general purpose computing device. Components of computing device 100 may include, but are not limited to, at least one processing unit 1001 as described above, at least one memory unit 1002 as described above, and a bus 1003 that connects the different system components (including processing unit 1001 and memory unit 1002).

Bus 1003 includes a data bus, a control bus, and an address bus.

The storage unit 1002 may include readable media in the form of volatile memory, such as Random Access Memory (RAM) 10021 and/or cache memory 10022, and may further include readable media in the form of non-volatile memory, such as Read Only Memory (ROM) 10023.

The storage unit 1002 may also include a program/utility 10025 having a set (at least one) of program modules 10024, such program modules 10024 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each or some combination of which may include an implementation of a network environment.

Computing device 100 may also communicate with one or more external devices 1004 (e.g., keyboard, pointing device, etc.). Such communication may occur through an input/output (I/O) interface 1005. Moreover, computing device 100 may also communicate with one or more networks such as a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network such as the Internet via network adapter 1006. As shown in fig. 10, the network adapter 1006 communicates with other modules of the computing device 100 over the bus 1003. It should be appreciated that although not shown, other hardware and/or software modules may be used in connection with computing device 100, including, but not limited to, microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, data backup storage systems, and the like.

It should be noted that although in the above detailed description several units/modules or sub-units/modules of the session answering means are mentioned, this division is only exemplary and not mandatory. Indeed, the features and functionality of two or more units/modules described above may be embodied in one unit/module in accordance with embodiments of the present disclosure. Conversely, the features and functions of one unit/module described above may be further divided into ones that are embodied by a plurality of units/modules.

Furthermore, although the operations of the methods of the present disclosure are depicted in the drawings in a particular order, this is not required or suggested that these operations must be performed in this particular order or that all of the illustrated operations must be performed in order to achieve desirable results. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step to perform, and/or one step decomposed into multiple steps to perform.

While the spirit and principles of the present disclosure have been described with reference to several particular embodiments, it is to be understood that this disclosure is not limited to the particular embodiments disclosed nor does it imply that features in these aspects are not to be combined to benefit from this division, which is done for convenience of description only. The disclosure is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A conversation answering method, comprising:

Get the first question;

Based on the first question, determine the target seat;

Obtaining a first historical reply from a historical interaction database based on the first question and the identity of the target agent, wherein the first historical reply corresponds to the first question and the identity of the target agent, and/or corresponds to a second question and the identity of the target agent, the second question being a similar question to the first question, and the semantic similarity between the similar question and the first question being greater than a preset semantic similarity threshold; wherein the historical interaction database is constructed based on historical questions, historical replies corresponding to the historical questions, and the identity of the human agent who processed the historical questions;

determining a response style of the target agent based on the first historical response;

Determining, based on the first question, a topic vector corresponding to the first question, and/or determining, based on the first question, context data corresponding to the first question;

generating, based on the topic vector and/or the context data, the response style, and the first question, a response content consistent with the target agent's language style;

The determining, according to the first question, a topic vector corresponding to the first question includes:

Obtaining, from the historical interaction database, a second historical answer according to the first question, wherein the second historical answer corresponds to the first question and/or corresponds to the second question;

Based on the second historical answers, a topic vector corresponding to the first question is determined.

2. The conversation answering method according to claim 1, wherein determining the target agent's answering style based on the first historical answer comprises:

Performing word segmentation processing on the first historical reply to obtain multiple word segmentation results;

Determining a part-of-speech sequence of the first historical reply based on the parts-of-speech of the multiple word segmentation results;

Based on the part-of-speech sequence, the answering style of the target agent is obtained.

3. The conversation answering method according to claim 1, further comprising:

Get the length of the second historical reply;

filtering the second historical replies according to a preset reply length range and the length of the second historical replies, wherein the preset reply length range is determined according to the mean and variance of the lengths of the second historical replies;

The determining, based on the second historical answer, a topic vector corresponding to the first question includes:

Based on the filtered second historical answers, a topic vector corresponding to the first question is determined.

4. The conversation answering method according to claim 1, wherein determining the topic vector corresponding to the first question based on the second historical reply comprises:

Perform topic extraction on the second historical answer to obtain a topic vector corresponding to the first question.

5. The conversation answering method according to claim 4, wherein extracting topics from the second historical answers to obtain a topic vector corresponding to the first question comprises:

Obtaining a first preset number of topics from the second historical answers, wherein each topic carries a corresponding weight and each topic includes a second preset number of phrases;

According to the weight corresponding to each topic, obtaining the topic with the largest weight from the first preset number of topics;

A topic vector corresponding to the first question is determined based on a second preset number of phrases included in the topic with the largest weight.

6. The conversation answering method according to claim 5, wherein determining the topic vector corresponding to the first question based on the second preset number of phrases included in the topic with the largest weight comprises:

performing vectorization processing on a second preset number of phrases included in the topic with the largest weight to obtain the second preset number of phrase vectors;

accumulating the second preset number of phrase vectors and calculating an average value;

The average value is used as a topic vector corresponding to the first question.

7. The conversation answering method according to claim 1 or 2, further comprising: after determining context data corresponding to the first question based on the first question;

splicing the context data according to the order of the context data;

If the number of words in the concatenated context data is greater than a third preset number, deleting the concatenated context data so that the number of words in the deleted context data is less than or equal to the third preset number;

Generating response content according to the response style and the first question includes:

Generate response content based on the topic vector and/or the abridged context data, the response style, and the first question.

8. The conversation answering method according to claim 1 or 2, wherein generating the answer content based on the topic vector and/or the context data, the answer style, and the first question comprises:

Performing vectorization processing on the context data, the answer style, and the first question respectively to obtain a context data vector, an answer style vector, and a first question vector;

The topic vector and/or the context data vector, as well as the answer style vector and the first question vector are fused, and the answer content is generated based on the fused vector.

9. The conversation answering method according to claim 8, wherein fusing the topic vector and/or the context data vector, the answer style vector, and the first question vector comprises:

The topic vector and/or the context data vector, the answer style vector, and the first question vector are concatenated together to obtain the fused vector.

10. The conversation answering method according to claim 1 or 2, before generating the answer content based on the topic vector and/or the context data, the answer style, and the first question, further comprising:

Obtaining the target agent's response to the first question;

Generating the response content according to the topic vector and/or the context data, the response style, and the first question includes:

Generate response content based on the responded content, the topic vector and/or the context data, the response style and the first question.

11. A computer-readable storage medium, wherein computer-executable instructions are stored in the computer-readable storage medium, and when a processor executes the computer-executable instructions, the conversation answering method according to any one of claims 1 to 10 is implemented.

12. A conversation answering device, comprising:

An acquisition module, used for acquiring a first question;

A first determination module, configured to determine a target agent based on the first question;

an obtaining module, configured to obtain the answering style of the target agent based on the first question and the identity of the target agent;

a generating module, configured to generate, based on the answering style and the first question, an answer content consistent with the language style of the target agent;

The obtaining module is specifically configured to obtain, from a historical interaction database, a first historical reply based on the first question and the identity of the target agent, wherein the first historical reply corresponds to the first question and the identity of the target agent, and/or corresponds to a second question and the identity of the target agent, the second question being a similar question to the first question, and the semantic similarity between the similar question and the first question being greater than a preset semantic similarity threshold; wherein the historical interaction database is constructed based on historical questions, historical replies corresponding to the historical questions, and the identity of a human agent who processed the historical questions; and determining the reply style of the target agent based on the first historical reply;

a second determining module, configured to determine, based on the first question, a topic vector corresponding to the first question, and/or, based on the first question, determine context data corresponding to the first question;

The generation module is specifically configured to generate a response content based on the topic vector and/or the context data, the response style, and the first question;

The second determination module is specifically used to obtain a second historical answer in the historical interaction database according to the first question, wherein the second historical answer corresponds to the first question and/or corresponds to the second question; based on the second historical answer, determine the topic vector corresponding to the first question.

13. The conversation response device according to claim 12, wherein the obtaining module is specifically configured to:

14. The conversation answering device according to claim 12, wherein the second determining module is specifically configured to:

Get the length of the second historical reply;

15. The conversation answering device according to claim 12, wherein the second determining module is specifically configured to:

16. The conversation answering device according to claim 15, wherein the second determining module is specifically configured to:

17. The conversation answering device according to claim 16, wherein the second determining module is specifically configured to:

18. The conversation answering device according to claim 12 or 13, wherein the second determining module is further configured to:

splicing the context data according to the order of the context data;

The generation module is specifically used to:

19. The conversation response device according to claim 12 or 13, wherein the generating module is specifically configured to:

20. The conversation response device according to claim 19, wherein the generating module is specifically configured to:

21. The conversation response device according to claim 12 or 13, wherein the generating module is specifically configured to:

Obtaining the target agent's response to the first question;

22. A computing device comprising: at least one processor and a memory;

The memory stores computer-executable instructions;

The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the conversation answering method according to any one of claims 1 to 10.