CN110688454A - Method, device, equipment and storage medium for processing consultation conversation - Google Patents

Abstract

The present application relates to the field of artificial intelligence, and provides a method, device, device and storage medium for processing consultation dialogue. Classify the product information to obtain a classification result; build a product information database according to the classification result; when receiving the consultation information input by the user, input the consultation information into the dialogue consultation model to obtain the consultation intention information of the user; The second product information corresponding to the consultation intention information is matched in the product knowledge base, and the semantic representation of the second product information is converted into a discourse language, and a consultation return result is generated and output; Statistics and analysis are performed on the results returned from historical consultation, and multi-dimensional reports are generated and output. By adopting this solution, the service efficiency of the enterprise service platform for product consultation can be improved.

Description

Method, device, device and storage medium for consulting dialogue processing

technical field

The present application relates to the field of intelligent decision-making, and in particular, to a method, apparatus, device and storage medium for consulting dialogue processing.

Background technique

In the current product sales management, the consultant generally sends a query request to the system processing center through the enterprise service platform, and the system processing center analyzes the information of the query request, obtains the consulting intention information of the consultant, and matches the corresponding information according to the consulting intention information. The consultation result is returned to the enterprise service platform, and the consultation result is displayed in the form of text on the enterprise service platform.

Because the accuracy of the consultation intention information obtained by the current enterprise service platform for the query request input by the consultant is low, the matching degree between the obtained consultation result and the consultation request is low, so that the consultant needs to input consultation information multiple times or In the end, the desired consultation result could not be obtained. Since the current enterprise service platform displays the consultation results in the form of text, the display method is fixed and single, so that it takes a relatively long time for the consultant to understand the consultation results displayed on the enterprise service platform. Therefore, the service efficiency of the enterprise service platform for product consultation is low.

SUMMARY OF THE INVENTION

The present application provides a method, device, device and storage medium for processing consultation dialogue, which can solve the problem of low service efficiency of product consultation by enterprise service platforms in the prior art.

In a first aspect, the present application provides a method for processing a consultation dialogue, the method comprising:

Obtain training information, input the training information into the model, and train the model to obtain a dialogue consultation model, which is used to analyze and process the target product information input by the user to construct product information database, which analyzes and processes the consultation information input by the user to obtain the returned result corresponding to the consultation information, wherein the training information includes a variety of first product information, consultation dialogue information and network knowledge information. The consulting model includes a first sub-model and a second sub-model;

Receive the target product information input by the user, classify the target product information through the dialogue consultation model, and obtain a classification result, wherein the classification result includes the attribute, field and production stage of the target product;

A first correspondence relationship and/or a second correspondence relationship is established according to the classification result, and a product information database is constructed according to the first correspondence relationship or the second correspondence relationship, wherein the first correspondence relationship includes target product information and The corresponding relationship of target information, the second corresponding relationship includes the corresponding relationship between classification information, product feature information and recommendation information;

When receiving the consultation information input by the user, input the consultation information into the first sub-model, and obtain key information from the consultation information through the first sub-model;

inputting the key information into the second sub-model, and analyzing the key information through the second sub-model to obtain the consultation intention information of the user;

Matching the second product information corresponding to the consultation intention information in the product knowledge base, converting the semantic representation of the second product information into a discourse language, and generating and outputting a consultation return result;

Perform statistics and analysis on historical key information and historical consultation results, and generate and output multi-dimensional reports, wherein the historical key information includes key information obtained from multiple consultation information received within a preset time period, and the The historical consultation return results include consultation return results obtained from multiple consultation information received within a preset time period.

In a possible design, the obtaining training information, inputting the training information into a model, and training the model to obtain a dialogue consultation model, include:

obtaining training information, inputting the training information into a neural network model, classifying the first product information and the network knowledge information according to preset association rules, respectively obtaining the first association information and the second association information, and Obtain the consultation purpose information of the consultation dialogue information;

Establish a correspondence between the consultation purpose information, the first associated information and the second associated information, and when the neural network model detects the consultation purpose information, output the first associated information and the second associated information second associated information, inputting the consultation purpose information, the corresponding relationship, the first associated information and the second associated information into the neural network model;

Multi-group parameter training is performed on the neural network model to obtain the neural network model to be trained under the multi-group parameters;

evaluating the error of the neural network model under each set of parameters to obtain a plurality of error values;

The minimum error value is obtained by calculating the size of a plurality of the error values, a set of parameters corresponding to the minimum error value is used as the parameters of the neural network model, the target neural network model is obtained, and the target neural network model is performed. Deploy, and obtain the dialogue consultation model, wherein the deployment refers to packaging, installing, configuring and publishing the configuration file, user manual, and help document of the application target neural network model.

In a possible design, the dialogue consultation model includes a regression analysis model, the first corresponding relationship and/or the second corresponding relationship is established according to the classification result, and the first corresponding relationship or the second corresponding relationship is established according to the classification result. Before the relationship builds the product information base, the method further includes:

Input analysis data into the dialogue consultation model, wherein the analysis data includes product user information and purchase information, product sales situation information, consumption levels and consumption directions in various places, and purchases in various places in the product field;

Regression analysis is performed on the analysis data through the regression analysis model to obtain recommendation information, wherein the recommendation information includes product development trends, developable regions, saleable groups, and products other than the analyzed products in the field to which the product belongs. information;

The establishing a first corresponding relationship and/or a second corresponding relationship according to the classification result, and constructing a product information database according to the first corresponding relationship or the second corresponding relationship, including:

Classify the target product information according to the categories of product functions, product types and product structures to obtain classification information;

obtaining the first feature information of the classification information;

A second correspondence between the categorization information, the first feature information and the recommendation information is established to obtain a product information database.

In a possible design, the dialogue consultation model includes a capsule network model, the key information is input into the second sub-model, and the key information is analyzed by the second sub-model to obtain the user's consultation. Intent information, including:

Perform at least one of the following text data processing methods on the key information to obtain target consultation information: domain analysis, text error correction, text completion, metaphor resolution, word decomposition, part-of-speech tagging, entity recognition, and text feature extraction;

Perform intent classification on the target consultation information through the capsule network model to obtain intent classification information;

Obtain attribute information in the intention classification information, and perform reasoning and contextual decision on the attribute information to obtain the user's consultation intention information.

In a possible design, before the text data processing is performed on the key information, the method further includes:

Obtaining training data, wherein the training data includes a statement query language and a query language;

Attribute analysis is performed on the training data, and the attribute-analyzed training data is marked to obtain words with marked attributes, wherein the attribute analysis includes proper noun attribute analysis, gender attribute analysis, singular and plural attribute analysis, distance attribute analysis Attribute analysis and abbreviated matching attribute analysis;

Obtain the second feature information of the training data, and select a feature from the second feature information as the splitting criterion of the node, wherein the calculation formula for obtaining the splitting criterion is as follows:

Among them, Info(D) is entropy, G(A) is the information gain rate, D is the training data, m is the number of words of the tag attribute, and p _i is a selected one of the second feature information The word of the marked attribute corresponding to the feature, A is one of the attributes in the marked attribute, and v is the number of outputs corresponding to the attribute A test and the number of divisions of the output;

Using the word of the marked attribute as the root node, construct a decision tree according to the preset classification standard and the splitting standard;

The performing text data processing on the key information includes:

Perform attribute analysis on the key information, and mark the key information subjected to the attribute analysis to obtain the target word of the marked attribute;

Perform data analysis and processing on the target word through the decision tree.

In a possible design, after the second product information corresponding to the consultation intention information is matched in the product knowledge base, the method further includes:

obtaining a matching degree, wherein the matching degree is the degree of correspondence between the product information in the product knowledge base and the consulting intent information;

detecting whether the matching degree is less than a first preset threshold;

When the detection result is yes, analyze the consultation information to obtain information to be solved, wherein the information to be solved includes the field, name, technical problem and problem type of the product;

Match consultation and answering personnel according to the to-be-solved information, and send the consultation information and the prompt information of the answering time limit to the information acquisition tool, so that the information acquisition tool generates a solution result according to the consultation information and the prompt information ;

The solution result received by the information acquisition tool is acquired, the solution information and the solution result are input into the dialogue consultation model, the dialogue consultation model is trained, and an updated dialogue consultation model is obtained.

In a possible design, after the second product information corresponding to the consultation intention information is matched in the product knowledge base, the method further includes:

acquiring a plurality of matching values, wherein the matching value is a corresponding degree value between the product information in the product knowledge base and the consulting intention information;

Calculate a plurality of the matching values, compare the sizes of the multiple matching values, and obtain the maximum matching value;

detecting whether the maximum matching value is less than a second preset threshold;

If yes, output target recommendation information, wherein the target recommendation information includes product information corresponding to the maximum matching value, recommendation information matched with the product information, and product information associated with the product information.

In a second aspect, the present application provides an apparatus for consultation dialogue processing, which has a function of implementing the method corresponding to the consultation dialogue processing provided in the first aspect. The functions can be implemented by hardware, or can be implemented by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions, and the modules may be software and/or hardware.

In a possible design, the device includes:

The input and output module is used to obtain training information, receive the target product information input by the user, and receive the consultation information input by the user;

The processing module is configured to input the training information obtained by the input and output module into the model, and train the model to obtain a dialogue consultation model, which is used to analyze and analyze the target product information input by the user. processing, to build a product information database, analyze and process the consulting information input by the user, to obtain a return result corresponding to the consulting information; Carry out classification and obtain classification results; establish a first correspondence and/or a second correspondence according to the classification results, and build a product information database according to the first correspondence or the second correspondence; The received consultation information is input into the first sub-model, and key information is obtained from the consultation information through the first sub-model; the key information is input into the second sub-model, through the The second sub-model analyzes the key information to obtain the user's consultation intention information; matches the second product information corresponding to the consultation intention information in the product knowledge base, and converts the semantic representation of the second product information Convert it into a vocabulary language, generate a consultation return result and input the consultation return result to the display module through the input and output module; perform statistics and analysis on historical key information and historical consultation return results, generate a multi-dimensional report and pass the The input and output module inputs the multi-dimensional report to the display module;

The display module is configured to receive and display the return result of the consultation and the multi-dimensional report form from the input and output module.

In a possible design, the processing module is specifically used for:

obtaining training information, inputting the training information into a neural network model, classifying the first product information and the network knowledge information according to preset association rules, respectively obtaining the first association information and the second association information, and Obtain the consultation purpose information of the consultation dialogue information;

Establish a correspondence between the consultation purpose information, the first associated information and the second associated information, and when the neural network model detects the consultation purpose information, output the first associated information and the second associated information second associated information, inputting the consultation purpose information, the corresponding relationship, the first associated information and the second associated information into the neural network model;

Multi-group parameter training is performed on the neural network model to obtain the neural network model to be trained under the multi-group parameters;

evaluating the error of the neural network model under each set of parameters to obtain a plurality of error values;

The minimum error value is obtained by calculating the size of a plurality of the error values, a set of parameters corresponding to the minimum error value is used as the parameters of the neural network model, the target neural network model is obtained, and the target neural network model is performed. Deploy, and obtain the dialogue consultation model, wherein the deployment refers to packaging, installing, configuring and publishing the configuration file, user manual, and help document of the application target neural network model.

In a possible design, the processing module includes a regression analysis model when executing the dialogue consultation model, and the first corresponding relationship and/or the second corresponding relationship is established according to the classification result, and the first corresponding relationship is established according to the first corresponding relationship. Or the second correspondence is also used for:

Input analysis data into the dialogue consultation model, wherein the analysis data includes product user information and purchase information, product sales situation information, consumption levels and consumption directions in various places, and purchases in various places in the product field;

Regression analysis is performed on the analysis data through the regression analysis model to obtain recommendation information, wherein the recommendation information includes product development trends, developable regions, saleable groups, and products other than the analyzed products in the field to which the product belongs. information;

The establishing a first corresponding relationship and/or a second corresponding relationship according to the classification result, and constructing a product information database according to the first corresponding relationship or the second corresponding relationship, including:

Classify the target product information according to the categories of product functions, product types and product structures to obtain classification information;

obtaining the first feature information of the classification information;

A second correspondence between the categorization information, the first feature information and the recommendation information is established to obtain a product information database.

In a possible design, the processing module is specifically used for:

Perform at least one of the following text data processing methods on the key information to obtain target consultation information: domain analysis, text error correction, text completion, metaphor resolution, word decomposition, part-of-speech tagging, entity recognition, and text feature extraction;

Perform intent classification on the target consultation information through the capsule network model to obtain intent classification information;

Obtain attribute information in the intention classification information, and perform reasoning and contextual decision on the attribute information to obtain the user's consultation intention information.

In a possible design, before executing the text data processing on the key information, the processing module is further configured to:

Obtaining training data, wherein the training data includes a statement query language and a query language;

Attribute analysis is performed on the training data, and the attribute-analyzed training data is marked to obtain words with marked attributes, wherein the attribute analysis includes proper noun attribute analysis, gender attribute analysis, singular and plural attribute analysis, distance attribute analysis Attribute analysis and abbreviated matching attribute analysis;

Obtain the second feature information of the training data, and select a feature from the second feature information as the splitting criterion of the node, wherein the calculation formula for obtaining the splitting criterion is as follows:

Among them, Info(D) is entropy, G(A) is the information gain rate, D is the training data, m is the number of words of the tag attribute, and p _i is a selected one of the second feature information The word of the marked attribute corresponding to the feature, A is one of the attributes in the marked attribute, and v is the number of outputs corresponding to the attribute A test and the number of divisions of the output;

Using the word of the marked attribute as the root node, construct a decision tree according to the preset classification standard and the splitting standard;

The performing text data processing on the key information includes:

Perform attribute analysis on the key information, and mark the key information subjected to the attribute analysis to obtain the target word of the marked attribute;

Perform data analysis and processing on the target word through the decision tree.

In a possible design, after performing the matching of the second product information corresponding to the consultation intention information in the product knowledge base, the processing module is further configured to:

obtaining a matching degree, wherein the matching degree is the degree of correspondence between the product information in the product knowledge base and the consulting intent information;

detecting whether the matching degree is less than a first preset threshold;

When the detection result is yes, analyze the consultation information to obtain information to be solved, wherein the information to be solved includes the field, name, technical problem and problem type of the product;

Match consultation and answering personnel according to the to-be-solved information, and send the consultation information and the prompt information of the answering time limit to the information acquisition tool, so that the information acquisition tool generates a solution result according to the consultation information and the prompt information ;

The solution result received by the information acquisition tool is acquired, the solution information and the solution result are input into the dialogue consultation model, the dialogue consultation model is trained, and an updated dialogue consultation model is obtained.

In a possible design, after performing the matching of the second product information corresponding to the consultation intention information in the product knowledge base, the processing module is further configured to:

acquiring a plurality of matching values, wherein the matching value is a corresponding degree value between the product information in the product knowledge base and the consulting intention information;

Calculate a plurality of the matching values, compare the sizes of the multiple matching values, and obtain the maximum matching value;

detecting whether the maximum matching value is less than a second preset threshold;

If so, output target recommendation information, wherein the target recommendation information includes product information corresponding to the maximum matching value, recommendation information matched with the product information, and product information associated with the product information.

Yet another aspect of the present application provides a computer device, comprising at least one connected processor, a memory, a display, and an input-output unit, wherein the memory is used to store program codes, and the processor is used to invoke the memory in the memory The program code to execute the method described in the first aspect.

Yet another aspect of the present application provides a computer storage medium, which includes instructions that, when executed on a computer, cause the computer to perform the method described in the first aspect above.

Compared with the prior art, in the solution provided by the present application, the target product information input by the user is received, the target product information is classified by the dialogue consultation model, and the classification result is obtained; the product information is constructed according to the classification result. database; when receiving the consultation information input by the user, input the consultation information into the dialogue consultation model to obtain the consultation intention information of the user; match the second consultation intention information corresponding to the consultation intention information in the product knowledge base product information, and convert the semantic representation of the second product information into a vocabulary language, generate and output consultation return results; perform statistics and analysis on historical key information and historical consultation return results, and generate and output multi-dimensional reports. Since the product information is organized and classified to build a product information database, the product information corresponding to the consulting information can be quickly and accurately obtained, and the product information associated with the consulting information can be obtained in the product information database. product information, and present the returned results of the consultation in multiple aspects and angles to improve the accuracy and versatility of the returned results of the consultation; display the returned results of the consultation in terms of language and multi-dimensional Understand the return result of the consultation from an angle, thereby shortening the time for the user to obtain the content of the return result of the consultation. By improving the accuracy and speed of obtaining the returned results of the consultation, and shortening the time for users to obtain the content of the returned results of the consultation, the number of consultation times and consultation time of the user to the enterprise service platform can be reduced, thereby improving the enterprise service platform's response to the consultation. Product consulting service efficiency.

Description of drawings

1 is a schematic flowchart of a method for processing a consultation dialogue in an embodiment of the application;

Fig. 2 is a kind of explanatory diagram of an example of the result of consultation return in the embodiment of the application;

FIG. 3 is an illustrative diagram of target product information in the embodiment of the application;

4 is a schematic structural diagram of a first sub-model in an embodiment of the present application;

FIG. 5 is another schematic structural diagram of the first sub-model in the embodiment of the present application;

6 is a schematic structural diagram of an apparatus for processing a consultation dialogue in an embodiment of the present application;

FIG. 7 is a schematic structural diagram of a computer device in an embodiment of the present application.

The realization, functional characteristics and advantages of the purpose of the present application will be further described with reference to the accompanying drawings in conjunction with the embodiments.

Detailed ways

It should be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application. The terms "first", "second" and the like in the description and claims of the present application and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It is to be understood that data so used may be interchanged under appropriate circumstances so that the embodiments described herein can be practiced in sequences other than those illustrated or described herein. Furthermore, the terms "comprising" and "having", and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device comprising a series of steps or modules is not necessarily limited to those expressly listed Those steps or modules, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products or devices, the division of modules appearing in this application is only a logical division , in practical applications, there may be other division methods, for example, multiple modules may be combined or integrated in another system, or some features may be ignored or not implemented.

The present application provides a method, device, device and storage medium for consulting dialog processing, which can be used for consulting management of products by an enterprise intelligence platform.

In order to solve the above-mentioned technical problems, the application mainly provides the following technical solutions:

By receiving the target product information input by the user, classifying the target product information through the dialogue consultation model to obtain a classification result; building a product information database according to the classification result; when receiving the consultation information input by the user, Input the consultation information into the dialogue consultation model to obtain the consultation intention information of the user; match the second product information corresponding to the consultation intention information in the product knowledge base, and express the semantics of the second product information The form is converted into a vocabulary language, and the consultation return results are generated and output; the historical key information and historical consultation return results are counted and analyzed, and multi-dimensional reports are generated and output. Since the product information is organized and classified to build a product information database, the product information corresponding to the consulting information can be quickly and accurately obtained, and the product information associated with the consulting information can be obtained in the product information database. product information, and present the returned results of the consultation in multiple aspects and angles to improve the accuracy and versatility of the returned results of the consultation; display the returned results of the consultation in terms of language and multi-dimensional Understand the return result of the consultation from an angle, thereby shortening the time for the user to obtain the content of the return result of the consultation. By improving the accuracy and speed of obtaining the returned results of the consultation, and shortening the time for users to obtain the content of the returned results of the consultation, the number of consultation times and consultation time of the user to the enterprise service platform can be reduced, thereby improving the enterprise service platform's response to the consultation. Product consulting service efficiency.

Referring to FIG. 1, the following provides an example of a method for processing a consultation dialogue provided by the present application, and the method includes:

101. Obtain training information, input the training information into the model, and train the model to obtain a dialogue consultation model.

Among them, the dialogue consultation model is used to analyze and process the target product information input by the user to build a product information database, and analyze and process the consultation information input by the user to obtain the return result corresponding to the consultation information; the training information includes multiple A first product information, consultation dialogue information and network knowledge information; the dialogue consultation model includes a first sub-model and a second sub-model.

The network knowledge information includes: first other product information other than the first product information that is the same and/or similar to the field to which the product of the first product information belongs, and is of the same type as the product of the product information /or similar second product information other than the first product information.

The dialogue consultation model is used to analyze and process the target product information input by the user to build a product information database, and analyze and process the consultation information input by the user to obtain the consultation return result corresponding to the consultation information. associated return results).

In some embodiments, the consultation return result includes the information of the product being consulted and the information of the recommended product with the same or similar field, type and specific performance as the product being consulted, as shown in FIG. 2 . The content in Figure 2 is for reference only, and its accuracy and actual operation are not considered.

102. Receive the target product information input by the user, classify the target product information through a dialogue consultation model, and obtain a classification result.

Among them, the classification results include the attributes, fields and production stages of the target product.

The target product information includes various types of information about the product, the person in charge of the product, and the workflow node information of the person in charge. The workflow node information includes the project formulation and decision-making of the product, the deployment of various work in the early stage of the product, and the work in the mid-term of the product. The deployment of the product and the deployment of various work in the later stage of the product. For example: when the user enters "Haofi APP product" in the terminal device, the dialogue consultation model will output product information such as the function, structure and performance related to the Haofi APP product, as well as the personal information of the product development team and the content it is responsible for ,As shown in Figure 3. The content in FIG. 3 is for reference only, and its accuracy and actual operation are not considered.

103. Establish a first correspondence relationship and/or a second correspondence relationship according to the classification result, and build a product information database according to the first correspondence relationship or the second correspondence relationship.

Wherein, the first corresponding relationship includes the corresponding relationship between target product information and target information, and the second corresponding relationship includes the corresponding relationship between classification information, product feature information and recommendation information.

In some embodiments, the dialogue consultation model includes a knowledge base system created from the collected massive network data, which can be used to build a product information base on the basis of the knowledge base system.

By constructing a product information base, the target information corresponding to the target product information can be quickly and accurately obtained according to the target product information.

104. When receiving the consultation information input by the user, input the consultation information into the first sub-model, and obtain key information from the consultation information through the first sub-model.

The first sub-model includes an image recognition sub-model, a text recognition sub-model and an audio and video recognition sub-model, and the first sub-model further includes a classifier. The image recognition sub-model, the text recognition sub-model and the audio and video recognition sub-model are connected in parallel, and the image recognition sub-model, the text recognition sub-model and the audio and video recognition sub-model are respectively connected in series with the classifier.

For example: the consultation information entered by the user in the APP is a product image, the consultation dialogue model receives the product image, and inputs the product image into the classifier, the classifier identifies the product image, and inputs the product image into the image recognition subclass In the model, the image recognition sub-model identifies and obtains the shape of the product in the product image, matches the corresponding product image in the product image database according to the obtained product shape, and obtains the product name corresponding to the product image, and uses the product name as consulting information. key information, as shown in Figure 4.

In some embodiments, the first sub-model further includes a plurality of filters, which are filter 1 in the image recognition sub-model, filter 2 in the text recognition sub-model, and filter 3 in the audio-video recognition sub-model. The image recognition sub-model, the text recognition sub-model and the audio and video recognition sub-model are connected in parallel, and the image recognition sub-model, the text recognition sub-model and the audio and video recognition sub-model are respectively connected in series with a filter.

For example: the consultation information entered by the user in the APP is a product picture, the consultation dialogue model receives the product picture, and inputs the product picture into the image recognition sub-model, text recognition sub-model and audio and video recognition sub-model, and the image recognition sub-model The respective filters in the model, text recognition sub-model and audio and video recognition sub-model recognize the product image. If the filter in the image recognition sub-model recognizes that the input information is a picture, the image recognition sub-model will continue to work and recognize Obtain the shape of the product in the product image, match the corresponding product image in the product image database according to the obtained product shape, and obtain the product name corresponding to the product image, and use the product name as the key information of the consultation information; if the text The filter in the recognition sub-model recognizes that the input information is non-text information, and the text recognition sub-model does not output; the same is true for the audio and video recognition sub-model. As shown in Figure 5.

In this application, product information can be consulted by means of online video and robot dialogue consultation, text input and picture input. It can be seen that this application can realize multi-channel input to improve user experience.

105. Input the key information into the second sub-model, and analyze the key information through the second sub-model to obtain the consultation intention information of the user.

The second sub-model is used to analyze and process key information to obtain the user's consulting intention information. The second sub-model analyzes the received key information, and obtains the consultation intention information of the user, and the obtained consultation intention information of the user may include various contents.

For example, the key information received by the second sub-model is "memory card, use", and the second sub-model analyzes the key information "memory card, use", analyzes and obtains the possible corresponding user's consultation intention information as "storage card, use" How to store the card", "What other products can be used with the memory card" and "The main issues of using the memory card and the related problems encountered during use and their solutions".

106. Match the second product information corresponding to the consultation intention information in the product knowledge base, convert the semantic representation of the second product information into a vocabulary language, and generate and output a consultation return result.

By converting the semantic representation of the second product information into a discourse language, the time required for the consultant to conceive the sales expression in the language of the second product information is reduced, thereby facilitating the rapid use of the second product information by the consultant.

107. Perform statistics and analysis on historical key information and historical consultation results, and generate and output multi-dimensional reports.

The historical key information includes the key information obtained from the received multiple consultation information within a preset time, and the historical consultation return result includes the consultation return result obtained from the multiple received consultation information within the preset time.

The multi-dimensional report includes at least the following three items: the number of users consulted, the type of consultation content, the number of consultations, the information of repeated consultations, the consultation information processed manually in the background, and the consultation information processed and unprocessed manually in the background.

Compared with the existing mechanism, in the embodiment of the present application, by receiving the target product information input by the user, the target product information is classified through the dialogue consultation model, and the classification result is obtained; the product information database is constructed according to the classification result; When the consultation information is received, the consultation information is input into the dialogue consultation model to obtain the consultation intention information of the user; the second product information corresponding to the consultation intention information is matched in the product knowledge base, and the semantic representation of the second product information is converted into It can generate and output the results of consultations using the terminology language; conduct statistics and analysis on historical key information and historical consultation results, and generate and output multi-dimensional reports. Because the product information is organized and classified to build a product information database, so as to obtain the product information corresponding to the consulting information quickly and accurately, and obtain the product information related to the product information corresponding to the consulting information in the product information database. Display the results of the consultation return from various aspects and multiple angles to improve the accuracy and versatility of the consultation return results; display the consultation return results in terms of language and multi-dimensional reports, so that users can understand the consultation return results from multiple angles, thereby shortening the user's response to consultation returns. The acquisition time of the result content. By improving the accuracy and speed of obtaining the results returned by the consultation, and shortening the time for users to obtain the content of the returned consultation results, users can reduce the number of consultations and consultation time for the enterprise service platform, thereby improving the enterprise service platform's product consultation. service efficiency.

Optionally, in some embodiments of the present application, the above-mentioned obtaining training information, inputting the training information into the model, and training the model to obtain the dialogue consultation model, include:

Obtain training information, input the training information into the neural network model, classify the first product information and network knowledge information respectively according to the preset association rules, obtain the first associated information and the second associated information, and obtain the consultation of consultation dialogue information purpose information;

Establish a corresponding relationship between the consultation purpose information, the first associated information and the second associated information. When the neural network model detects the consultation purpose information, it outputs the first associated information and the second associated information, and the consultation purpose information, the corresponding relationship, the first associated information an association information and a second association information are input into the neural network model;

Perform multiple sets of parameter training on the neural network model to obtain the neural network model to be trained under multiple sets of parameters;

Evaluate the error of the neural network model under each set of parameters to obtain multiple error values;

The minimum error value is obtained by calculating the size of multiple error values, a set of parameters corresponding to the minimum error value is used as the parameters of the neural network model, the target neural network model is obtained, the target neural network model is deployed, and the dialogue consultation model is obtained, wherein , deployment refers to packaging, installing, configuring and publishing the configuration files, user manuals, and help documents of the application target neural network model.

Among them, the neural network model is composed of input nodes, output nodes and node layers. The parameters include at least one of the size of the convolution kernel, the learning rate η, the batch parameter, the number of neural network layers, the activation function, the optimizer, the batch size, and the number of hyperparameter epochs for training.

The accuracy of the neural network model is improved by evaluating the error of the neural network model to be trained under multiple sets of parameters. By deploying the error-evaluated neural network model, a dialogue consultation model with relatively complete functions and relatively stable performance can be obtained.

Optionally, in some embodiments of the present application, the above-mentioned dialogue consultation model includes a regression analysis model, a first corresponding relationship and/or a second corresponding relationship is established according to the classification result, and a first corresponding relationship or a second corresponding relationship is established according to the classification result. Before the product information base, methods include:

Input analysis data into the dialogue consultation model, wherein the analysis data includes product user information and purchase information, product sales situation information, consumption level and consumption direction of various places, and purchase situation of various places in the product field;

Perform regression analysis on the analysis data through a regression analysis model to obtain recommendation information, where the recommendation information includes product development trends, developable regions, saleable groups, and product information other than the analysis products in the field to which the product belongs;

Establishing a first correspondence and/or a second correspondence according to the classification results, and building a product information database according to the first correspondence or the second correspondence, including:

Classify the target product information according to the categories of product functions, product types and product structures to obtain classification information;

obtaining the first feature information of the classification information;

A second correspondence between the classification information, the first feature information and the recommendation information is established to obtain a product information database.

By performing regression analysis on the analysis data including product user information and purchase information, product sales information, consumption level and consumption direction in various places, and analysis data of the purchase situation in various parts of the product field, the recommendation information of multi-category content is obtained to enhance the The content of the results returned by the consultation of the dialogue consultation module is multifaceted, thereby improving the practicability and service efficiency of the enterprise service platform corresponding to the dialogue consultation module.

By first classifying the target product information to obtain the classification information, and then establishing the corresponding relationship of the classification information, the product information database is created, so that the information classification of the product information database is refined, thereby improving the efficiency of obtaining information in the product information database. and accuracy.

Optionally, in some embodiments of the present application, the above-mentioned dialogue consultation model includes a capsule network model, and the key information is input into the second sub-model, and the key information is analyzed by the second sub-model to obtain the user's consultation intention information, including: :

Perform at least one of the following text data processing methods on key information to obtain target consultation information: domain analysis, text error correction, text completion, metaphor elimination, word decomposition, part-of-speech tagging, entity recognition, and text feature extraction;

Intent classification is performed on the target consultation information through the capsule network model, so as to facilitate the intention classification information;

Obtain the attribute information in the intent classification information, and perform reasoning and contextual decision on the attribute information to obtain the user's consulting intent information.

Because the capsule network model has the characteristics of less training data required and better classification and identification, and through contextual decision-making to make the obtained information coherent, the obtained information is more accurate. Therefore, the capsule network model is used to consult the target. The information is classified into intent, and reasoning and contextual decision-making are performed according to the intent classification information, so as to improve the accuracy of obtaining the user's consulting intent information.

Optionally, in some embodiments of the present application, before the above-mentioned text data processing is performed on the key information, the method further includes:

Obtain training data, wherein the training data includes statement query language and query language;

Attribute analysis is performed on the training data, and the attribute-analyzed training data is marked to obtain words with marked attributes. The attribute analysis includes proper noun attribute analysis, gender attribute analysis, singular and plural attribute analysis, distance attribute analysis and abbreviation attribute analysis. Slightly matching attribute analysis;

Obtain the second feature information of the training data, and select a feature from the second feature information as the splitting criterion of the node, wherein the calculation formula for obtaining the splitting criterion is as follows:

Among them, Info(D) is the entropy, G(A) is the information gain rate, D is the training data, m is the number of words with tag attributes, and p _i is the tag attribute corresponding to a feature selected in the second feature information. word, A is one of the attributes in the tag attribute, v is the number of outputs corresponding to the attribute A test and the number of output divisions;

Using the word that marks the attribute as the root node, build a decision tree according to the preset classification criteria and split criteria;

Text data processing of key information, including:

Perform attribute analysis on the key information, mark the key information after the attribute analysis, and obtain the target word of the marked attribute;

The target word is analyzed and processed through the decision tree.

Attributes are selected by the information gain rate to overcome the deficiencies of selecting attributes with more values when using information gain to select attributes, so as to perform pruning in the process of tree construction to complete the discretization of continuous attributes, so that incomplete data can be processed by pruning. to be processed.

A decision tree is constructed by taking the word with marked attributes as the root node and according to the preset classification criteria and splitting criteria, so that the information of each node of the constructed decision tree corresponding to the target word is multi-angle and comprehensive. The accuracy of the corresponding user's consultation intention information.

Optionally, in some embodiments of the present application, after the above-mentioned second product information corresponding to the consultation intention information is matched in the product knowledge base, the method further includes:

Obtain the matching degree, where the matching degree is the degree of correspondence between the product information in the product knowledge base and the consulting intent information;

detecting whether the matching degree is less than a first preset threshold;

When the detection result is yes, analyze the consultation information to obtain the information to be solved, wherein the information to be solved includes the field, name, technical problem and problem type of the product;

Match the consulting and answering personnel according to the information to be solved, and send the consulting information and the prompt information of the answering time limit to the information acquisition tool, so that the information acquisition tool can generate a solution result according to the consulting information and prompt information;

Obtain the solution result received by the information acquisition tool, input the solution information and the solution result into the dialogue consultation model, train the dialogue consultation model, and obtain the updated dialogue consultation model.

The matching degree between the product information in the product knowledge base and the consultation intention information is compared with the first preset threshold, so as to return it to the relevant staff when the relevance of the information obtained from the product information in the product knowledge base is not high, so as to Obtain more accurate information and train the solution results returned by the obtained staff. When the relevant information to be solved is identified again, the dialogue consultation model automatically obtains and outputs the corresponding consultation return results. Through this operation, the accuracy of the returned result of the consultation and the practicability of the dialogue consultation model are improved, thereby improving the service efficiency of the enterprise service platform corresponding to the dialogue consultation model.

For example, the consultation information is "No response to the wealth management APP", the consultation intention information is "Solution corresponding to the failure of the wealth management APP to analyze and process the information", and the first preset threshold is "not less than 75%". According to the “solution corresponding to the “financial management APP does not respond to the analysis and processing of information”, the matching degree of the information obtained from the consulting intention information in the product information in the product knowledge base is 40%, and 40% is less than 75%, then the consulting information is analyzed. , in order to obtain the pending information of “How to solve the financial management APP if there is no response to the analysis and processing of information” and “Service management platform-financial management APP-fault problem-no response to the background operation-how to operate”, and the pending information and The prompt information for answering the time limit is sent to the relevant person in charge of the technical department in the form of email. After receiving the solution result returned by the relevant person in charge of the technical department, the solution information and the solution result are input into the dialogue consultation model, and the dialogue consultation model is trained and obtained. Updated conversational consultation model. When the dialogue consultation model detects the same or similar information to be solved again, it automatically obtains and outputs the solution result.

Optionally, in some embodiments of the present application, after the above-mentioned second product information corresponding to the consultation intention information is matched in the product knowledge base, the method further includes:

Obtaining a plurality of matching values, wherein the matching value is the corresponding degree value between the product information in the product knowledge base and the consulting intention information;

Calculate multiple matching values, compare the sizes of multiple matching values, and obtain the largest matching value;

detecting whether the maximum matching value is less than a second preset threshold;

If so, output target recommendation information, wherein the target recommendation information includes product information corresponding to the maximum matching value, recommendation information matched with the product information, and product information associated with the product information.

By comparing the maximum matching value among the plurality of product information related to the consultation intention information obtained in the product knowledge base according to the consultation intention information and the size of the second preset threshold, the target recommendation result can be output correspondingly according to the judgment result, and the consultation intention can be improved. The accuracy or relevance of the returned results is convenient for the consultant to obtain the information of the returned results of the consultation and reduce the number of consultation dialogues with the enterprise service platform, thereby improving the service efficiency of the enterprise service platform corresponding to the dialogue consultation model.

Optionally, in some embodiments of the present application, the above-mentioned evaluation of the error of the neural network model under each set of parameters to obtain a plurality of error values, including:

Obtain the test value and theoretical value of the neural network model under each set of parameters;

Calculate the test value and theoretical value, and obtain the error of the neural network model under each set of parameters to obtain multiple error values. The calculation formula for obtaining multiple error values is as follows:

Among them, H(p,q) is the cross-entropy cost function, p(x) is the theoretical value, q(x) is the test value, and N is the number of samples.

Wherein, the test value refers to the actual accuracy rate of the returned results corresponding to the consultation dialogue information output in the test set, and the theoretical value refers to the expected accuracy rate of the output results corresponding to the consultation dialogue information in the test set.

The error value is calculated by calculating the cross-entropy cost function to measure how good the prediction of the neural network model is.

Optionally, in some embodiments of the present application, the above-mentioned obtaining training information, inputting the training information into the model, and training the model to obtain the dialogue consultation model, include:

Mark the technical fields, product types, product attributes and product description types of a variety of first product information, and mark the technical fields, product types, product attributes and product description types of network knowledge information;

According to the marked content, extract the first feature information of the first product information and the second feature information of the network knowledge information, and respectively convert the semantic representations of the first feature information and the second feature information into semantics according to product sales terms Representation form to obtain the first characteristic speech information and the second characteristic speech information;

Obtain the keywords of the consultation dialogue information, and establish a corresponding relationship between the keywords, the first characteristic speech information and the second characteristic speech information, wherein the correspondence is used to obtain the first characteristic speech information when the keyword is detected. with the second characteristic discourse information;

Through the keywords, the first characteristic speech information, the second characteristic speech information and the corresponding relationship, the dialogue consultation model satisfies the preset rule conditions, so as to obtain the corresponding consultation return results, wherein the preset rule conditions include the input of the user. The target product information is sorted and analyzed to build a product information database and obtain corresponding product information by analyzing the consulting information input by the user.

Optionally, in some embodiments of the present application, the dialogue consultation model includes an information base that has been created, and the first correspondence and/or the second correspondence is established according to the classification result, and the first correspondence is established according to the first correspondence. Or the second corresponding relationship builds a product information base, including:

Obtain the product type, product field and product attribute of the target product information through the dialogue consultation model;

Obtain the related product information corresponding to the product type and the field from the information database, and use the related product information as the target information;

The first correspondence between the target product information and the target information is established through the dialogue consultation model, the target product information is classified according to the product attributes, the classification information is obtained, and the product information database is constructed according to the classification information and the first correspondence.

For example, if the target product information is trust wealth management information, the trust wealth management information is analyzed through the dialogue consultation model, and the obtained product type is the corporate wealth management service type, the product field is the financial field, and the product attributes include interpretation, income type and various types of trust wealth management. Corresponding features, the trust financial information is classified according to the income type to obtain classification information. From the information database, we obtained the information of ICBC which is quite competitive with corporate wealth management service and the information of ICBC's corporate wealth management service products, and classified ICBC's corporate wealth management service products as fixed income. Category, Guaranteed Category and Non Guaranteed Floating Income Category, and obtain various types of characteristic information (such as the content of the product and the type of people it is suitable for). Establish the first correspondence between trust wealth management information and ICBC's corporate wealth management service product information. When the dialogue consultation model receives the input "trust wealth management", the dialogue consultation model outputs various information of trust wealth management, and outputs various information of ICBC's corporate wealth management service products in the form of recommended information.

The technical features mentioned in any of the embodiments or implementation manners in the above-mentioned FIGS. 1 to 5 are also applicable to the embodiments corresponding to FIGS. 6 and 7 in the present application, and the similarities will not be repeated hereafter.

A method for consultation dialogue processing in the present application is described above, and the device 60 for executing the above consultation dialogue processing method is described below.

As shown in FIG. 6, a schematic structural diagram of an apparatus 60 for processing consultation dialogue can be applied to consultation management of products by an enterprise intelligence platform. The apparatus 60 in this embodiment of the present application can implement steps corresponding to the method for processing a consultation dialogue executed in any of the embodiments or implementations in FIG. 1 to FIG. 5 above. The functions implemented by the apparatus 60 may be implemented by hardware, or may be implemented by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions, and the modules may be software and/or hardware. The device 60 may include an input/output module 601, a processing module 602 and a display module 603, and the function implementation of the input/output module 601, the processing module 602 and the display module 603 may refer to any embodiment or implementation in FIG. 1 to FIG. 5 . The operations performed in the mode are not repeated here. The processing module 602 can be used to control the input and output operations of the input and output module 601 , and the display module 603 can be used to display the processing operations of the processing module 602 .

In some embodiments, the input and output module 601 can be used to obtain training information, receive target product information input by the user, and receive consultation information input by the user;

The processing module 602 is used to input the training information obtained by the input and output module 601 into the model, and train the model to obtain a dialogue consultation model. The dialogue consultation model is used to analyze and process the target product information input by the user to construct a The product information database analyzes and processes the consultation information input by the user to obtain the return result corresponding to the consultation information; classifies the target product information received by the input and output module 601 through the dialogue consultation model, and obtains the classification result; establishes a classification result according to the classification result The first correspondence and/or the second correspondence, according to the first correspondence or the second correspondence, construct the product information database; Input the consultation information received by the input and output module 601 into the first sub-model, and through the first sub-model from the Obtain key information from the consultation information; input the key information into the second sub-model, and analyze the key information through the second sub-model to obtain the user's consultation intention information; match the second product information corresponding to the consultation intention information in the product knowledge base, Convert the semantic representation of the second product information into a vocabulary language, generate a consultation return result and input the consultation return result to the display module 603 through the input and output module 601; perform statistics and analysis on historical key information and historical consultation return results, Generate a multi-dimensional report and input the multi-dimensional report to the display module 603 through the input and output module 601;

The display module 603 is configured to receive and display the consultation return result and the multi-dimensional report form from the input and output module 601 .

The training information includes a variety of first product information, consultation dialogue information and network knowledge information. The dialogue consultation model includes a first sub-model and a second sub-model.

The network knowledge information includes: first other product information other than the first product information that is the same and/or similar to the field to which the product of the first product information belongs, and is of the same type as the product of the product information /or similar second product information other than the first product information.

The dialogue consultation model is used to analyze and process the target product information input by the user to build a product information database, and analyze and process the consultation information input by the user to obtain the consultation return result corresponding to the consultation information. associated return results).

In some embodiments, the consultation return result includes the information of the product being consulted and the information of the recommended product with the same or similar field, type and specific performance as the product being consulted.

Among them, the classification results include the attributes, fields and production stages of the target product.

The target product information includes various types of information about the product, the person in charge of the product, and the workflow node information of the person in charge. The workflow node information includes the project formulation and decision-making of the product, the deployment of various work in the early stage of the product, and the work in the mid-term of the product. The deployment of the product and the deployment of various work in the later stage of the product. For example: when the user enters "Haofi APP product" in the terminal device, the dialogue consultation model will output product information such as the function, structure and performance related to the Haofi APP product, as well as the personal information of the product development team and the content it is responsible for .

Wherein, the first corresponding relationship includes the corresponding relationship between target product information and target information, and the second corresponding relationship includes the corresponding relationship between classification information, product feature information and recommendation information.

In some embodiments, the dialogue consultation model includes a knowledge base system created from the collected massive network data, which can be used to build a product information base on the basis of the knowledge base system.

By constructing a product information base, the target information corresponding to the target product information can be quickly and accurately obtained according to the target product information.

The first sub-model includes an image recognition sub-model, a text recognition sub-model and an audio and video recognition sub-model, and the first sub-model further includes a classifier. The image recognition sub-model, the text recognition sub-model and the audio and video recognition sub-model are connected in parallel, and the image recognition sub-model, the text recognition sub-model and the audio and video recognition sub-model are respectively connected in series with the classifier.

In some embodiments, the first sub-model further includes a plurality of filters, which are filter 1 in the image recognition sub-model, filter 2 in the text recognition sub-model, and filter 3 in the audio-video recognition sub-model. The image recognition sub-model, the text recognition sub-model and the audio and video recognition sub-model are connected in parallel, and the image recognition sub-model, the text recognition sub-model and the audio and video recognition sub-model are respectively connected in series with a filter.

The second sub-model is used to analyze and process key information to obtain the user's consulting intention information. The second sub-model analyzes the received key information, and obtains the consultation intention information of the user, and the obtained consultation intention information of the user may include various contents.

By converting the semantic representation of the second product information into a discourse language, the time required for the consultant to conceive the sales expression in the language of the second product information is reduced, thereby facilitating the rapid use of the second product information by the consultant.

The historical key information includes the key information obtained from the received multiple consultation information within a preset time, and the historical consultation return result includes the consultation return result obtained from the multiple received consultation information within the preset time.

The multi-dimensional report includes at least the following three items: the number of users consulted, the type of consultation content, the number of consultations, the information of repeated consultations, the consultation information processed manually in the background, and the consultation information processed and unprocessed manually in the background.

In the embodiment of the present application, the processing module 602 classifies the target product information through the dialogue consultation model through the target product information input by the user received by the input and output module 601, and obtains the classification result; builds a product information database according to the classification result; When the consultation information input by the user obtained by the output module 601 is input, the consultation information is input into the dialogue consultation model to obtain the consultation intention information of the user; the second product information corresponding to the consultation intention information is matched in the product knowledge base, and the second product The semantic representation of the information is converted into a vocabulary language, and the consultation return result is generated and output through the display module 603 ; Because the product information database is constructed by arranging and classifying the product information, so as to obtain the product information corresponding to the consultation information quickly and accurately, and obtain the product information associated with the product information corresponding to the consultation information in the product information database, Display the results of consultation in multiple aspects and angles to improve the accuracy and versatility of the results of consultation; display the results of consultation in terms of language and multi-dimensional reports, so that users can understand the results of consultation from multiple perspectives, thereby shortening the user's understanding of the results. Inquire about the acquisition time of the returned result content. By improving the accuracy and speed of obtaining the returned results of the consultation, and shortening the time for users to obtain the content of the returned results of the consultation, the number of consultation times and consultation time of the user to the enterprise service platform can be reduced, thereby improving the enterprise service platform's response to the consultation. Product consulting service efficiency.

Optionally, in some embodiments of the present application, any embodiment or technical features mentioned in the above-mentioned method for processing a consultation dialogue are also applicable to the method for executing the above-mentioned consultation dialogue processing in this application. The device 60 is not repeated hereafter.

The device 60 in the embodiment of the present application is respectively introduced from the perspective of modular functional entities above, and a computer device is introduced from the perspective of hardware below, as shown in FIG. 7 , which includes: a processor, a memory, a display, an input and output unit ( Also a transceiver, not identified in Figure 7) and a computer program stored in the memory and executable on the processor. For example, the computer program may be a program corresponding to the method for processing a consultation dialogue in any of the embodiments or implementations of FIG. 1 to FIG. 5 . For example, when the computer device implements the function of the device 60 as shown in FIG. 6 , when the processor executes the computer program, the processor implements the method for consulting dialog processing performed by the device 60 in the embodiment corresponding to FIG. 6 above. Each step; or, when the processor executes the computer program, the functions of each module in the apparatus 60 of the embodiment corresponding to FIG. 6 are implemented. For another example, the computer program may be a program corresponding to the method for processing a consultation dialogue in any of the embodiments or implementations in FIG. 1 to FIG. 5 .

The processor may be a central processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf processors Programmable Gate Array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the computer device, and uses various interfaces and lines to connect various parts of the entire computer device.

The memory can be used to store the computer program and/or module, and the processor implements the computer by running or executing the computer program and/or module stored in the memory and calling the data stored in the memory various functions of the device. The memory may mainly include a stored program area and a stored data area, wherein the stored program area may store an operating system, an application program required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the storage data area may store Data (such as audio data, video data, etc.) created according to the usage of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory such as hard disk, internal memory, plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card , a flash memory card (Flash Card), at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device.

The input and output units can also be replaced by receivers and transmitters, which can be the same or different physical entities. When they are the same physical entity, they can be collectively referred to as input and output units. The input-output unit may be a transceiver.

The memory may be integrated in the processor, or may be provided separately from the processor.

From the description of the above embodiments, those skilled in the art can clearly understand that the method of the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is better implementation. Based on this understanding, the technical solutions of the present application can be embodied in the form of software products in essence or the parts that make contributions to the prior art. The computer software products are stored in a storage medium (such as ROM/RAM), including Several instructions are used to cause a terminal (which may be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of this application.

The embodiments of the present application have been described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific embodiments, which are merely illustrative rather than restrictive. Under the inspiration of this application, without departing from the scope of protection of the purpose of this application and the claims, many forms can be made. Directly or indirectly applied in other related technical fields, these all fall within the protection of this application.

Claims (10)

1. A method for consulting dialogue processing, characterized in that the method comprises: Obtain training information, input the training information into the model, and train the model to obtain a dialogue consultation model, which is used to analyze and process the target product information input by the user to construct product information database, which analyzes and processes the consultation information input by the user to obtain the returned result corresponding to the consultation information, wherein the training information includes a variety of first product information, consultation dialogue information and network knowledge information. The consulting model includes a first sub-model and a second sub-model; receiving the target product information input by the user, and classifying the target product information through the dialogue consultation model to obtain a classification result, wherein the classification result includes the attribute, field and production stage of the target product; A first correspondence relationship and/or a second correspondence relationship is established according to the classification result, and a product information database is constructed according to the first correspondence relationship and/or the second correspondence relationship, wherein the first correspondence relationship includes a target product The correspondence between information and target information, and the second correspondence includes the correspondence between classification information, product feature information and recommendation information; When receiving the consultation information input by the user, input the consultation information into the first sub-model, and obtain key information from the consultation information through the first sub-model; inputting the key information into the second sub-model, and analyzing the key information through the second sub-model to obtain the consultation intention information of the user; Matching the second product information corresponding to the consultation intention information in the product knowledge base, converting the semantic representation of the second product information into a discourse language, and generating and outputting a consultation return result; Perform statistics and analysis on historical key information and historical consultation results, and generate and output multi-dimensional reports, wherein the historical key information includes key information obtained from multiple consultation information received within a preset time period, and the The historical consultation return results include consultation return results obtained from multiple consultation information received within a preset time period. 2. The method according to claim 1, wherein the obtaining training information, inputting the training information into a model, and training the model to obtain a dialogue consultation model, comprising: obtaining training information, inputting the training information into a neural network model, classifying the first product information and the network knowledge information according to preset association rules, respectively obtaining the first association information and the second association information, and Obtain the consultation purpose information of the consultation dialogue information; Establish a correspondence between the consultation purpose information, the first associated information and the second associated information, and when the neural network model detects the consultation purpose information, output the first associated information and the second associated information second associated information, inputting the consultation purpose information, the corresponding relationship, the first associated information and the second associated information into the neural network model; Multi-group parameter training is performed on the neural network model to obtain the neural network model to be trained under the multi-group parameters; evaluating the error of the neural network model under each set of parameters to obtain a plurality of error values; The minimum error value is obtained by calculating the size of a plurality of the error values, a set of parameters corresponding to the minimum error value is used as the parameters of the neural network model, the target neural network model is obtained, and the target neural network model is performed. Deploy, and obtain the dialogue consultation model, wherein the deployment refers to packaging, installing, configuring and publishing the configuration file, user manual, and help document of the application target neural network model. 3. The method according to claim 1, wherein the dialogue consultation model comprises a regression analysis model, and the establishment of a first correspondence and/or a second correspondence according to the classification result, according to the first Before the corresponding relationship or the second corresponding relationship constructs the product information database, the method further includes: Input analysis data into the dialogue consultation model, wherein the analysis data includes product user information and purchase information, product sales situation information, consumption levels and consumption directions in various places, and purchases in various places in the product field; Regression analysis is performed on the analysis data through the regression analysis model to obtain recommendation information, wherein the recommendation information includes product development trends, developable regions, saleable groups, and products other than the analyzed products in the field to which the product belongs. information; The establishing a first corresponding relationship and/or a second corresponding relationship according to the classification result, and constructing a product information database according to the first corresponding relationship or the second corresponding relationship, including: Classify the target product information according to the categories of product functions, product types and product structures to obtain classification information; obtaining the first feature information of the classification information; A second correspondence between the categorization information, the first feature information and the recommendation information is established to obtain a product information database. 4. The method according to claim 1, wherein the dialogue consultation model comprises a capsule network model, the key information is input into the second sub-model, and the second sub-model analyzes all the data. key information to obtain the user's consultation intention information, including: Perform at least one of the following text data processing methods on the key information to obtain target consultation information: domain analysis, text error correction, text completion, metaphor resolution, word decomposition, part-of-speech tagging, entity recognition, and text feature extraction; Perform intent classification on the target consultation information through the capsule network model to obtain intent classification information; Obtain attribute information in the intention classification information, and perform reasoning and contextual decision on the attribute information to obtain the user's consultation intention information. 5. The method according to claim 4, wherein before the text data processing is performed on the key information, the method further comprises: Obtaining training data, wherein the training data includes a statement query language and a query language; Attribute analysis is performed on the training data, and the attribute-analyzed training data is marked to obtain words with marked attributes, wherein the attribute analysis includes proper noun attribute analysis, gender attribute analysis, singular and plural attribute analysis, distance attribute analysis Attribute analysis and abbreviated matching attribute analysis; Obtain the second feature information of the training data, and select a feature from the second feature information as the splitting criterion of the node, wherein the calculation formula for obtaining the splitting criterion is as follows:

Among them, Info(D) is entropy, G(A) is the information gain rate, D is the training data, m is the number of words of the tag attribute, and p _i is a selected one of the second feature information The word of the marked attribute corresponding to the feature, A is one of the attributes in the marked attribute, and v is the number of outputs corresponding to the attribute A test and the number of divisions of the output; Using the word of the marked attribute as the root node, construct a decision tree according to the preset classification standard and the splitting standard; The performing text data processing on the key information includes: Attribute analysis is performed on the key information, and the key information subjected to the attribute analysis is marked to obtain the target word of the marked attribute; Perform data analysis and processing on the target word through the decision tree. 6 . The method according to claim 1 , wherein after matching the second product information corresponding to the consultation intention information in the product knowledge base, the method further comprises: 6 . obtaining a matching degree, wherein the matching degree is the degree of correspondence between the product information in the product knowledge base and the consulting intent information; detecting whether the matching degree is less than a first preset threshold; When the detection result is yes, analyze the consultation information to obtain information to be solved, wherein the information to be solved includes the field, name, technical problem and problem type of the product; Match consultation and answering personnel according to the to-be-solved information, and send the consultation information and the prompt information of the answering time limit to the information acquisition tool, so that the information acquisition tool generates a solution result according to the consultation information and the prompt information ; The solution result received by the information acquisition tool is acquired, the solution information and the solution result are input into the dialogue consultation model, the dialogue consultation model is trained, and an updated dialogue consultation model is obtained. 7 . The method according to claim 1 , wherein after matching the second product information corresponding to the consultation intention information in the product knowledge base, the method further comprises: 8 . acquiring a plurality of matching values, wherein the matching value is a corresponding degree value between the product information in the product knowledge base and the consulting intention information; Calculate a plurality of the matching values, compare the sizes of the multiple matching values, and obtain the maximum matching value; detecting whether the maximum matching value is less than a second preset threshold; If yes, output target recommendation information, wherein the target recommendation information includes product information corresponding to the maximum matching value, recommendation information matched with the product information, and product information associated with the product information. 8. A device for consulting dialogue processing, wherein the device comprises: The input and output module is used to obtain training information, receive the target product information input by the user, and receive the consultation information input by the user; The processing module is configured to input the training information obtained by the input and output module into the model, and train the model to obtain a dialogue consultation model, which is used to analyze and analyze the target product information input by the user. processing, to build a product information database, analyze and process the consulting information input by the user, to obtain a return result corresponding to the consulting information; Carry out classification and obtain classification results; establish a first correspondence and/or a second correspondence according to the classification results, and build a product information database according to the first correspondence or the second correspondence; The received consultation information is input into the first sub-model, and key information is obtained from the consultation information through the first sub-model; the key information is input into the second sub-model, through the The second sub-model analyzes the key information to obtain the consultation intention information of the user; matches the second product information corresponding to the consultation intention information in the product knowledge base, and converts the semantic representation of the second product information Convert it into a language, generate a consultation return result and input the consultation return result to the display module through the input and output module; perform statistics and analysis on historical key information and historical consultation return results, generate a multi-dimensional report and pass the The input and output module inputs the multi-dimensional report to the display module; The display module is configured to receive and display the return result of the consultation and the multi-dimensional report form from the input and output module. 9. A computer device, characterized in that the computer device comprises: at least one processor, memory, display, and input-output unit; Wherein, the memory is used for storing program codes, and the processor is used for calling the program codes stored in the memory to execute the method according to any one of claims 1-7. 10. A computer storage medium comprising instructions which, when run on a computer, cause the computer to perform the method of any of claims 1-7.

Images