CN113886563A

CN113886563A - Question recommendation method and device, electronic equipment and storage medium

Info

Publication number: CN113886563A
Application number: CN202111243349.4A
Authority: CN
Inventors: 彭程; 赵筱军; 张强
Original assignee: Beijing Dajia Internet Information Technology Co Ltd
Current assignee: Beijing Dajia Internet Information Technology Co Ltd
Priority date: 2021-10-25
Filing date: 2021-10-25
Publication date: 2022-01-04

Abstract

The disclosure relates to a question recommendation method and device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence. The method comprises the following steps: determining a user account identifier of a problem to be recommended; acquiring user portrait information and historical service information corresponding to a user account identifier and a service line identifier corresponding to a historical operation page; according to the user portrait information, the historical service information, the service line identification corresponding to the historical operation page and a pre-trained intention classification model, determining question intention classification information corresponding to the user account identification; and determining a target recommendation question according to the questioning intention classification information. The method and the device at least solve the problem that the question-asking intention of the user cannot be accurately predicted when the question is recommended to the user in the related technology, so that the accuracy of the recommendation question is poor.

Description

Question recommendation method and device, electronic equipment and storage medium

Technical Field

The present disclosure relates to the field of artificial intelligence technologies, and in particular, to a problem recommendation method and apparatus, an electronic device, and a storage medium.

Background

The intelligent customer service is widely applied as a service scene in which artificial intelligence is first to fall on the ground. Guessing the question is the first service provided by the intelligent customer service to the user, and guessing the accuracy of the recommendation problem directly influences the use experience of the user.

In the related art, a recommendation problem to a user account is determined by counting high-frequency problems in history reply problems. When the method is adopted to recommend the questions to the user, the question-asking intention of the user cannot be accurately predicted, so that the accuracy of the recommendation questions is poor.

Disclosure of Invention

The disclosure provides a question recommendation method, a question recommendation device, an electronic device and a storage medium, which are used for solving at least the problem that when a question is recommended to a user in the related art, the question intention of the user cannot be accurately predicted, and the accuracy of the recommendation question is poor. The technical scheme of the disclosure is as follows:

according to a first aspect of the embodiments of the present disclosure, there is provided a question recommendation method, including: determining a user account identifier to be subjected to problem recommendation; acquiring user portrait information and historical service information corresponding to a user account identifier and a service line identifier corresponding to a historical operation page; according to the user portrait information, the historical service information, the service line identification corresponding to the historical operation page and a pre-trained intention classification model, determining question intention classification information corresponding to the user account identification; and determining a target recommendation question according to the questioning intention classification information.

In one possible implementation, the step of determining the target recommendation question according to the questioning intention classification information comprises the following steps: determining a target recommendation question according to the question intention classification information and the historical question information; the historical question information includes historical question questions for a plurality of sample user accounts.

In another possible implementation, the questioning intention classification information includes service line classification information and question information corresponding to the service line, and the determining of the target recommended question according to the questioning intention classification information and the historical questioning information includes: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining the historical question belonging to the service line in the historical question information as a second question to be recommended; determining a target recommendation question according to the first question to be recommended, the second question to be recommended and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the questioning intention classification information includes service line classification information, and the determining of the target recommendation question according to the questioning intention classification information and the historical questioning information includes: determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining a target question to be recommended according to a historical question belonging to a service line in the historical question information and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the questioning intention classification information includes question information corresponding to the service line, and the determining of the target recommended question according to the questioning intention classification information and the historical questioning information includes: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining a target recommendation question according to the first question to be recommended and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the classifying information according to the questioning intention and determining the target recommendation question includes: determining a target recommendation question according to the questioning intention classification information, the historical questioning information and the preset question information; the preset question information is determined according to the current hotspot event information, and the historical question information comprises historical question questions of a plurality of sample user accounts.

In another possible implementation, the questioning intention classification information includes service line classification information and question information corresponding to the service line, and the determining of the target recommended question according to the questioning intention classification information, the historical questioning information and the preset question information includes: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining the historical question belonging to the service line in the historical question information as a second question to be recommended; determining a target recommendation problem according to the first to-be-recommended problem, the second to-be-recommended problem, a high-frequency problem in the historical question information and preset problem information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation manner, the questioning intention classification information includes service line category information, and the determining of the target recommended question according to the questioning intention classification information, the historical questioning information and the preset question information includes: determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining the historical question belonging to the service line in the historical question information as a second question to be recommended; determining a target recommendation problem according to the second to-be-recommended problem, the high-frequency problem in the historical question information and preset problem information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation manner, the questioning intention classification information includes question information corresponding to the service line, and the determining of the target recommended question according to the questioning intention classification information, the historical questioning information and the preset question information includes: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining a target recommendation problem according to the first to-be-recommended problem, the high-frequency problem in the historical question information and preset problem information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the classifying information according to the questioning intention and determining the target recommendation question includes: determining a target recommendation question according to the questioning intention classification information and preset question information; and the preset problem information is determined according to the current hotspot event information.

In another possible implementation, the questioning intention classification information includes question information corresponding to the service line, and the determining of the target recommended question according to the questioning intention classification information and preset question information includes: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; and determining a target recommendation problem according to the first to-be-recommended problem and preset problem information.

In another possible embodiment, the intention classification model is pre-trained by: acquiring user portrait information, historical service information and a service line identifier corresponding to a historical operation page of each sample user account; performing multi-task training on the classification neural network model through user portrait information, historical service information and service line identification corresponding to a historical operation page until the classification neural network model is converged to obtain an intention classification model; the multi-task training comprises task training for identifying the class of a business line to be asked of a sample user account and task training for identifying the information of the question to be asked of the sample user account.

According to a second aspect of the embodiments of the present disclosure, there is provided an issue recommendation apparatus including: the determining module is configured to execute the step of determining the user account identifier to be subjected to question recommendation; the acquisition module is configured to execute the acquisition of user portrait information and historical service information corresponding to the user account identifier and a service line identifier corresponding to the historical operation page; the classification module is configured to execute the question intention classification information corresponding to the user account identification according to the user portrait information, the historical service information, the service line identification corresponding to the historical operation page and a pre-trained intention classification model; the determination module is further configured to perform: and determining a target recommendation question according to the questioning intention classification information.

In a possible implementation, the determining module is specifically configured to perform: determining a target recommendation question according to the question intention classification information and the historical question information; the historical question information includes historical question questions for a plurality of sample user accounts.

In another possible implementation, the questioning intention classification information includes service line classification information and question information corresponding to the service line, and the determining module is specifically configured to perform: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining the historical question belonging to the service line in the historical question information as a second question to be recommended; determining a target recommendation question according to the first question to be recommended, the second question to be recommended and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the questioning intention classification information includes service line category information, and the determining module is specifically configured to perform: determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining a target question to be recommended according to a historical question belonging to a service line in the historical question information and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the questioning intention classification information includes question information corresponding to the service line, and the determining module is specifically configured to perform: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining a target recommendation question according to the first question to be recommended and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the determining module is specifically configured to perform: determining a target recommendation question according to the questioning intention classification information, the historical questioning information and the preset question information; the preset question information is determined according to the current hotspot event information, and the historical question information comprises historical question questions of a plurality of sample user accounts.

In another possible implementation, the questioning intention classification information includes service line classification information and question information corresponding to the service line, and the determining module is specifically configured to perform: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining the historical question belonging to the service line in the historical question information as a second question to be recommended; determining a target recommendation problem according to the first to-be-recommended problem, the second to-be-recommended problem, a high-frequency problem in the historical question information and preset problem information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the questioning intention classification information includes service line category information, and the determining module is specifically configured to perform: determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining the historical question belonging to the service line in the historical question information as a second question to be recommended; determining a target recommendation problem according to the second to-be-recommended problem, the high-frequency problem in the historical question information and preset problem information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the questioning intention classification information includes question information corresponding to the service line, and the determining module is specifically configured to perform: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining a target recommendation problem according to the first to-be-recommended problem, the high-frequency problem in the historical question information and preset problem information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the determining module is specifically configured to perform: determining a target recommendation question according to the questioning intention classification information and preset question information; and the preset problem information is determined according to the current hotspot event information.

In another possible implementation, the questioning intention classification information includes question information corresponding to the service line, and the determining module is specifically configured to perform: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; and determining a target recommendation problem according to the first to-be-recommended problem and preset problem information.

In another possible embodiment, the apparatus further includes a training module configured to perform: acquiring user portrait information, historical service information and a service line identifier corresponding to a historical operation page of each sample user account; performing multi-task training on the classification neural network model through user portrait information, historical service information and service line identification corresponding to a historical operation page until the classification neural network model is converged to obtain an intention classification model; the multi-task training comprises task training for identifying the class of a business line to be asked of a sample user account and task training for identifying the information of the question to be asked of the sample user account.

According to a third aspect of the embodiments of the present disclosure, there is provided an electronic apparatus including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the problem recommendation method of the first aspect and any of its possible embodiments described above.

According to a fourth aspect of embodiments of the present disclosure, there is provided a computer-readable storage medium, wherein instructions of the computer-readable storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the problem recommendation method of any one of the above-mentioned first aspects and any one of its possible implementations.

According to a fifth aspect of embodiments of the present disclosure, there is provided a computer program product comprising computer instructions which, when run on an electronic device, cause the electronic device to perform the problem recommendation method of the first aspect and any of its possible implementations.

The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects: the method comprises the steps of obtaining portrait information of a user, historical business information and business line identifications corresponding to historical operation pages, achieving basic coverage of all characteristic data which cause customer complaints of the user, enabling an intention classification model to fuse the portrait information of the user, the historical business information and the business line identifications corresponding to the historical operation pages to obtain questioning intention classification information, being capable of accurately predicting the questioning intention of the user, achieving determination of target recommendation questions according to the questioning intention classification information, being capable of accurately covering the questions which are proposed by the user intention, achieving the purpose of not asking for a first answer, further improving accuracy of questions recommended to the user, and improving use experience of the user.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure and are not to be construed as limiting the disclosure.

FIG. 1 is a flow diagram illustrating a method of question recommendation in accordance with an exemplary embodiment;

FIG. 2 is a flow diagram illustrating another problem recommendation method in accordance with an exemplary embodiment;

FIG. 3 is a flow diagram illustrating another problem recommendation method in accordance with an exemplary embodiment;

FIG. 4 is a flow diagram illustrating another problem recommendation method in accordance with an exemplary embodiment;

FIG. 5 is a flow diagram illustrating another problem recommendation method in accordance with an exemplary embodiment;

FIG. 6 is a flow diagram illustrating another problem recommendation method in accordance with an illustrative embodiment;

FIG. 7 is a flow diagram illustrating another problem recommendation method in accordance with an illustrative embodiment;

FIG. 8 is a schematic diagram illustrating an intent classification model in accordance with an exemplary embodiment;

FIG. 9 is a block diagram illustrating an issue recommendation device in accordance with an exemplary embodiment;

FIG. 10 is a block diagram illustrating an electronic device in accordance with an example embodiment.

Detailed Description

In order to make the technical solutions of the present disclosure better understood by those of ordinary skill in the art, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein are capable of operation in sequences other than those illustrated or otherwise described herein. The implementations described in the exemplary embodiments below are not intended to represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

Before describing the problem recommendation method provided by the present disclosure in detail, the application scenario and implementation environment related to the present disclosure are briefly described.

First, a brief description is given of an application scenario to which the present disclosure relates.

Aiming at the problems, the method for recommending the questions is provided, the user portrait information, the historical service information and the service line identification corresponding to the historical operation page are obtained, all characteristic data which cause customer complaints of the user are basically covered, the obtained question intention classification information can be obtained by fusing the user portrait information, the historical service information and the service line identification corresponding to the historical operation page through an intention classification model, the question intention of the user can be accurately predicted, the target recommendation questions can be determined according to the question intention classification information, the questions which are intentionally proposed by the user can be accurately covered, the question which is not asked and answered in advance is achieved, the accuracy of the questions recommended to the user is improved, and the use experience of the user is improved.

Next, the following briefly describes an implementation environment (implementation architecture) related to the present disclosure.

The problem recommendation method provided by the embodiment of the disclosure can be applied to electronic equipment. The electronic device may be a terminal device or a server. The terminal device can be a smart phone, a tablet computer, a palm computer, a vehicle-mounted terminal, a desktop computer, a notebook computer and the like. The server may be any one server or server cluster, and the disclosure is not limited thereto.

In addition, the information related to the user account identifier according to the present disclosure (including, but not limited to, user portrait information corresponding to the user account, historical business information, and business line identifier corresponding to the historical operation page) is information that is authorized by the user account or is sufficiently authorized by each party.

For the sake of understanding, the problem recommendation method provided by the present disclosure is specifically described below with reference to the accompanying drawings.

FIG. 1 is a flow diagram illustrating a question recommendation method for an electronic device, according to an example embodiment. As shown in fig. 1, the question recommendation method includes the following steps.

In S101, a user account identifier to be subjected to question recommendation is determined.

In one embodiment, the terminal device is pre-configured with an application. After the user account triggers the intelligent customer service of the application program, the service of guessing your question of the intelligent customer service is started, and the intelligent customer service sends a recommendation problem to the user account through a chat interface with the user account so that the user account can select a problem which the user wants to consult from the recommendation problems.

Optionally, the identifier of the user account triggering the intelligent customer service may be determined as the identifier of the user account to be subjected to question recommendation.

In one embodiment, the electronic device is a terminal device. And the terminal equipment determines the user account identifier to be subjected to problem recommendation based on the operation for triggering the intelligent customer service by the account. Illustratively, the terminal device determines an identifier of a user account triggering the intelligent customer service as a user account identifier to be recommended for the problem.

In another embodiment, the electronic device is a server. After the user account triggers the intelligent customer service of the application program, the terminal equipment sends a recommendation problem confirmation request to the server, the recommendation problem confirmation request comprises a user account identifier, and the server determines the user account identifier as the user account identifier to be subjected to problem recommendation. Illustratively, the recommendation question confirmation request includes an identification of the user account, which is an identification of the user account that triggers the intelligent customer service.

In S102, user portrait information corresponding to the user account identifier, historical service operation information, and a service line identifier corresponding to the historical operation page are obtained.

Optionally, user portrait information corresponding to the user account identifier is obtained from a local cache of the electronic device.

In one embodiment, the electronic device updates user representation information corresponding to each user account identifier in the application program by day granularity and stores the user representation information in a local cache so that the electronic device can be called when determining the recommendation problem.

Illustratively, the user profile information includes one or more of age, gender, new and old users, places of daily use, registration duration, fan count, attention count, and the like.

Optionally, the historical service information of the user account identifier is obtained, and the historical service information of the user account identifier on each service line can be obtained by reading the service line interface in real time.

In one embodiment, the historical service information may be service operation information within a preset time period. For example, it may be business operation information in the last month.

Illustratively, the business operation information may include viewing duration, order information, reporting information, live information, and the like. It can be understood that the viewing duration and the live broadcast information are service operation information on a live broadcast service line, the order information is service operation information on an electric service line, and the reporting information is service operation information on a safety service line.

Optionally, the service line identifier corresponding to the historical operation page of the user account identifier is obtained, and the service line identifier corresponding to the historical operation page of the user account identifier can be obtained by reading the service line interface in real time. The historical operation page is used for representing the pages which are historically browsed.

In an embodiment, if the user account identifier browses the live broadcast page, the service line identifier corresponding to the historical operation page is the identifier of the live broadcast service line. And if the user account identifier browses the e-commerce page, the service line identifier corresponding to the historical operation page is the e-commerce service line identifier. And if the user account identifier browses the search page, the service line identifier corresponding to the historical operation page is the identifier of the search service line.

In step S103, question intention classification information corresponding to the user account identifier is determined according to the user portrait information, the historical business information, the business line identifier corresponding to the historical operation page, and the pre-trained intention classification model.

Due to the user portrait information, the historical service information and the service line identifications corresponding to the historical operation pages, all feature data which possibly cause customer complaints of the users can be basically covered, so that the intention classification model fuses the user portrait information, the historical service information and the question-asking intention classification information output after the service line identifications corresponding to the historical operation pages are fused, and the question-asking intention of the users can be accurately predicted.

Optionally, the user portrait information, the historical service information, and the service line identifier corresponding to the historical operation page are used as input information of an intention classification model, and the intention classification model outputs question intention classification information corresponding to the user account identifier.

In one embodiment, after the intent classification model obtains the user portrait information, the user portrait information is normalized to use a numerical characterization of 0 to 1 for each portrait feature in the user portrait information.

In one embodiment, after the intention classification model acquires the historical service information, normalization processing is performed on each service operation feature in the historical service information, so that each service operation feature in the historical service information is represented by a value between 0 and 1.

In one embodiment, after the intention classification model obtains the service line identifiers corresponding to the historical operation pages, each service line identifier is converted into a target vector.

Optionally, the intention classification model obtains and outputs the questioning intention classification information through the normalized user portrait information, the normalized historical business information and the normalized target vector.

It should be noted that the question intention classification information includes a question map of the user account identifier and a confidence level of the question intention. The confidence of the questioning intention is used for representing the accuracy of the questioning intention, that is, the higher the value corresponding to the confidence is, the higher the accuracy of the questioning intention is.

In S104, the target recommended question is determined based on the questioning intention classification information.

In one embodiment, after obtaining the questioning intention classification information, the electronic device obtains the target recommendation question according to the questioning intention information predicted by the questioning intention classification information.

The questioning intention classification information output by the intention classification model can accurately predict the questioning intention of the user, so that the target recommended questions obtained by the electronic equipment based on the questioning intention classification information can accurately cover the questions intentionally provided by the user, and the questions are not asked and answered first, thereby improving the accuracy of the recommended questions and improving the use experience of the user.

Optionally, when the electronic device is a terminal device, after the terminal device determines the target recommendation question, the target recommendation question is pushed to the user through a chat interface of the application program, so that the user can select a question that the user intends to ask. It can be understood that the application is the application that initiates the "guess you want" service of the intelligent customer service in S101.

Optionally, when the electronic device is a server, after determining the target recommendation problem, the server sends the target recommendation problem to a terminal device corresponding to the application program, and pushes the target recommendation problem to the user through a chat interface of the application program, so that the user can select a problem that the user intends to ask a question. It can be understood that the application is the application that initiates the "guess you want" service of the intelligent customer service in S101.

In the embodiment, all characteristic data causing customer complaints of users are basically covered by acquiring the portrait information of the users, the historical service information and the service line identifications corresponding to the historical operation pages, so that the question classification information obtained by fusing the portrait information of the users, the historical service information and the service line identifications corresponding to the historical operation pages by the intention classification model can accurately predict the question intentions of the users, a target recommendation question can be determined according to the question intention classification information, the question intended by the users can be accurately covered, question answering is not carried out, the accuracy of the question recommended to the users is improved, and the use experience of the users is improved.

In one possible implementation, in conjunction with fig. 1, as shown in fig. 2, S104 includes S104 a.

In S104a, the target recommended question is determined based on the question intention classification information and the history question information.

The historical question information comprises historical question questions of a plurality of sample user accounts.

Optionally, after obtaining the questioning intention classification information, the electronic device obtains the target recommendation question according to the questioning intention information and the historical questioning information predicted by the questioning intention classification information.

Optionally, the historical question information is acquired in real time, the real-time performance of the questions in the historical question information is high by using the acquired historical question information, and when an emergency occurs on the platform, the historical question information can cover the questions corresponding to the emergency, so that the target recommendation questions can more accurately cover the question intentions of the user, and further, the quick response can be realized.

In one embodiment, the historical question information is obtained by actively asking questions within a preset time period through a user account of the application program.

In one embodiment, after the user account identification to be subjected to question recommendation is determined, or after the intent classification model outputs the question intent classification information, the historical question information is obtained in real time.

In the embodiment, the target recommendation problem is obtained by combining the historical question information acquired in real time, the problem that the user lacks characteristic data during cold start is solved, so that the question intention of a newly registered user account can be preset under the condition that the user lacks user portrait information, historical service information or service line identification corresponding to a historical operation page, a proper target recommendation problem is pushed to the user account, and the use experience of the user is improved. Furthermore, the historical question information is acquired in real time, so that the real-time performance of the historical question of the sample user account in the historical question information is high, the historical question information can accurately cover the question corresponding to the emergency event, the question of the user account can be quickly responded when the emergency event occurs on the platform, the target recommendation question can accurately cover the question which is intended to be provided by the user account, the question which is not asked and answered first is achieved, the accuracy of the question recommended to the user account is improved, and the use experience of the user is improved.

In one possible embodiment, the questioning intention classification information includes service line category information and question information corresponding to the service lines. In conjunction with FIG. 2, as shown in FIG. 3, S104a includes S104a1-S104a 4.

In S104a1, a first question to be recommended is determined according to a question in the question information whose confidence level is greater than a first preset threshold.

Optionally, the question information corresponding to the service line is used to characterize a specific question that the user intends to consult.

In one embodiment, the intent classification model is pre-trained by a Multi-task learning (MTL) structure. By training the multi-task learning structure on the intention classification model, the intention classification model can simultaneously execute two tasks, and the accuracy of each task can be improved. The first task is used for identifying service line information of user intention consultation, and the second task is used for identifying specific problems of user intention consultation.

In one embodiment, the issue information corresponding to the line of business includes a plurality of issues and a confidence level for each issue. The service line in the question information corresponding to the service line is the service line that the user intends to consult.

For example, the service line category information may be determined first, and then the problem information corresponding to each service line in the service line category information may be determined, so as to obtain the problem information corresponding to the service line.

Illustratively, the question information corresponding to the business line includes questions intended to be asked by the user, such as inquiry of logistics progress, inquiry of account status, return of goods and money, how to get hot, and the confidence of each question. The confidence coefficient is used for representing the accuracy of the corresponding question being the user intention consultation question, that is, the higher the confidence coefficient is, the higher the possibility that the question corresponding to the confidence coefficient is the user intention consultation question is.

Alternatively, the confidence may be characterized by a value between 0 and 1. The first preset threshold is 0.9.

It should be noted that the value of the first preset threshold may be other values according to actual situations, and the disclosure is not limited herein.

In one embodiment, the question information includes a plurality of questions and a confidence level for each of the plurality of questions. The electronic equipment determines the problem with the confidence coefficient larger than a first preset threshold value as a first problem to be recommended. For example, a question with a confidence level greater than 0.9 is determined as the first question to be recommended.

In one embodiment, the question information includes a plurality of questions and a confidence level for each of the plurality of questions. And sequencing the plurality of problems according to the confidence degrees to obtain a problem information queue. And determining the problems in the problem information queue sorted before the preset value as the first to-be-recommended problem. For example, the question ordered in the top five of the question information queue may be determined as the first question to be recommended. It should be noted that the top five includes the fifth question ordered in the question information queue.

The problem information corresponding to the service line represents the specific problem of the user intention consultation, so that the problem of which the confidence coefficient is greater than a first preset threshold value in the problem information is determined as a first to-be-recommended problem, the first to-be-recommended problem can more accurately cover the problem of the user intention question, the accuracy of the target recommendation problem can be higher when the target recommendation problem is determined according to the first to-be-recommended problem, and the use experience of the user is improved.

In S104a2, the service line with the confidence level greater than the second preset threshold value in the service line type information is determined.

Optionally, the service line category information is used to characterize service line information that the user intends to consult.

Optionally, the service line category information includes a plurality of service line identifications and a confidence level of each of the plurality of service line identifications.

Illustratively, the service line category information includes identifications of service lines such as live broadcast, e-commerce, commercialization, search, recommendation, security, etc., and a confidence corresponding to the identification of each service line. The confidence coefficient is used for representing the accuracy of the service line information that the corresponding service line is intended to be consulted by the user, that is, the higher the confidence coefficient is, the higher the possibility that the service line corresponding to the confidence coefficient is the service line information that the user is intended to consult is.

In one embodiment, the service lines with confidence level greater than a second preset threshold in the service line category information are determined, for example, the service lines with confidence level greater than the second preset threshold include live service lines and e-commerce service lines.

Alternatively, the confidence may be characterized by a value between 0 and 1.

Optionally, the second preset threshold is 0.9. It should be noted that the value of the second preset threshold may be other values according to actual situations, and the disclosure is not limited herein.

In S104a3, the historical question questions belonging to the service line in the historical question information are determined as the second question to be recommended.

Optionally, the historical question questions belonging to the service line determined at S104a2 in the historical question information, for example, the historical question questions belonging to the live service line and the e-commerce service line, are obtained and determined as the second question to be recommended.

The service line type information represents the service line information of the user intention consultation, the service line information of the user intention question is determined firstly, then the question corresponding to the service line information of the user intention question is selected from the historical question information according to the service line information of the user intention question, and the second question to be recommended is obtained, so that the second question to be recommended can cover the question of the user intention question more accurately, the accuracy of the target recommendation question obtained based on the second question to be recommended is improved, and the use experience of the user is improved.

In S104a4, a target recommendation question is determined according to the first question to be recommended, the second question to be recommended, and the high-frequency question in the history question information.

The high-frequency question is a question of which the question frequency in the historical question information meets a preset condition. For example, the question that satisfies the preset condition may be a question that the number of times the user actively asks satisfies a threshold.

In one embodiment, the questions in the historical question information, of which the number of active questions meets a threshold value, are determined as high-frequency questions. For example, a question in the history question information, which is asked by the user for more than 100 times actively, may be determined as a high-frequency question. The threshold value may be adjusted according to practical situations, which is not limited by the present disclosure.

Illustratively, the questions in the historical question information are sorted according to the number of questions asked of the sample user account, so as to obtain a historical question queue. And determining the questions sorted before the preset value in the historical question queue as high-frequency questions. For example, the top five questions in the historical question queue may be determined to be high frequency questions. It should be noted that the top five includes the fifth-ranked question in the historical question queue.

Optionally, when obtaining the first question to be recommended, the second question to be recommended and the high-frequency question in the historical question information, the electronic device writes the first question to be recommended, the second question to be recommended and the high-frequency question in the historical question information into the target result list respectively. And then, the electronic equipment reads the problems meeting the preset conditions from the target result list and determines the problems as target recommendation problems.

In one embodiment, the questions in the target result list are sorted according to the order in which the questions obtained in different ways are written into the target result list, so as to obtain a target result queue. It can be understood that the first question to be recommended, the second question to be recommended and the high-frequency question in the historical question information are obtained through different ways.

Optionally, the order in which the questions obtained in different ways are written into the target result list is determined according to the accuracy of the question prediction user's question intention obtained in different ways. The problem obtained by the way with high prediction accuracy is firstly written into the target result list, the problem written into the target result list is arranged at the front, the problem arranged at the front is firstly read, and the problem is sent to the user account to be subjected to the problem.

In one embodiment, the accuracy of the predicted user question intention of the first question to be recommended is higher than the accuracy of the predicted user question intention of the second question to be recommended, so that the first question written into the target result list is the first question to be recommended, and the second question written into the target result list is the second question to be recommended. At this time, each question in the first question to be recommended is ranked before the second question to be recommended.

Illustratively, when the target recommendation problem is determined, the problem which is sorted in the target result queue before the preset value is determined as the target recommendation problem. Wherein the preset value can be determined according to the number of questions pushed to the user. For example, each time 5 questions are recommended to the user, the preset value is 5, and at this time, the question ranked in the top five of the target result queue is determined as the target recommendation question. It should be noted that the top five here includes the fifth problem of ordering in the target result queue.

Alternatively, the terminal device may display the target recommendation question by turning the page, that is, by displaying the pages.

For example, when the terminal device displays the target recommendation question by turning pages, the question written into the target result list first may be displayed on the first page, and the question written into the target result list may be displayed on the subsequent page. For example, a first question to be recommended is displayed on a first page, and a second question to be recommended is displayed on a second page.

It should be noted that when the number of the first to-be-recommended questions or the second to-be-recommended questions is greater than the number that can be displayed on the page, the questions ranked in the top are preferentially displayed. For example, for the first question to be recommended, a question with high confidence level in the first question to be recommended is preferentially displayed. And preferentially displaying the questions with more question times in the second questions to be recommended aiming at the second questions to be recommended.

In the embodiment, the question with the confidence level greater than the first preset threshold value in the question information is determined as the first question to be recommended, so that the first question to be recommended can more accurately cover the question intention of the user. And under the condition that the confidence coefficient of the service line category information is greater than a second preset threshold value, selecting a problem corresponding to the service line information of the user intention problem from the historical question information to obtain a second problem to be recommended, so that the second problem to be recommended can more accurately cover the problem of the user intention problem. Furthermore, the target recommendation problem obtained based on the first to-be-recommended problem and the second to-be-recommended problem can be used for predicting the question intention of the user more accurately, so that the accuracy of the target recommendation problem selected as the question by the user is improved, and the use experience of the user is improved.

In a possible implementation manner, in a case that the questioning intention classification information includes service line classification information and question information corresponding to a service line, and there is no service line whose confidence coefficient is greater than a second preset threshold in the service line classification information, determining a target recommended question according to the questioning intention classification information and the historical questioning information, includes: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining a target recommendation question according to the first question to be recommended and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets a preset condition.

In an embodiment, when there is no service line whose confidence is greater than the second preset threshold in the service line category information, that is, the confidence corresponding to the service line identifier in the service line category information is less than the second preset threshold, this indicates that the possibility that the service line identifier in the service line category information is the service line information that the user intends to consult is low. At the moment, the electronic equipment directly selects a high-frequency question from the historical question information, and determines a target recommendation question according to the first question to be recommended and the high-frequency question. In one possible implementation, the questioning intent classification information includes line of business category information. According to the questioning intention classification information and the historical questioning information, determining a target recommendation question, which comprises the following steps: determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining a target question to be recommended according to a historical question belonging to a service line in the historical question information and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In the embodiment, the target recommendation problem is obtained by selecting the problem corresponding to the service line information of the user intention problem from the historical question information under the condition that the service line with the confidence coefficient larger than the second preset threshold exists in the service line category information, so that the target recommendation problem can more accurately cover the problem of the user intention, the accuracy of the target recommendation problem meeting the requirement of the question is improved, and the use experience of the user is improved. And under the condition that no service line with the confidence coefficient larger than a second preset threshold exists in the service line category information, selecting a high-frequency question in the historical question information as a target recommendation question, and improving the accuracy that the target recommendation question meets the question-asking intention of the user, so that the accuracy that the target recommendation question is selected as the question by the user is improved, and the use experience of the user is further improved.

In one possible embodiment, the questioning intent classification information includes question information corresponding to a line of business. According to the questioning intention classification information and the historical questioning information, determining a target recommendation question, which comprises the following steps: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining a target recommendation question according to the first question to be recommended and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In the embodiment, the question with the confidence level greater than the first preset threshold value in the question information is determined as the first question to be recommended, so that the first question to be recommended can more accurately cover the question intention of the user. Furthermore, due to the high-frequency questions in the historical question information, the accuracy of meeting the user question intention is higher, the target recommendation question is determined by combining the high-frequency questions and the first to-be-recommended question, the question intention of the user can be predicted more accurately by the target recommendation question, the accuracy of the target recommendation question selected as the question by the user is improved, and the use experience of the user is improved.

In one possible implementation, in conjunction with fig. 1, as shown in fig. 4, S104 includes S104 b.

In S104b, a target recommended question is determined based on the questioning intention classification information, the historical questioning information and the preset question information.

The preset question information is determined according to the current hotspot event information, and the historical question information comprises historical question questions of a plurality of sample user accounts.

Optionally, after obtaining the questioning intention classification information, the electronic device obtains the target recommended question according to the questioning intention information, the historical questioning information and the preset question information predicted by the questioning intention classification information.

In one embodiment, after the user account identification of the question to be recommended is determined, or after the question intention classification information is output by the intention classification model, the historical question information is obtained in real time.

Optionally, when a hot event occurs, the preset problem information corresponding to the hot event is manually set. And the electronic equipment acquires the preset problem information and stores the preset problem information in a local cache. It is understood that the hot event here refers to a hot event occurring with the platform that initiated the "guess you want".

For example, the hot event may be an event related to a service on the platform, such as that a live video cannot be viewed, cannot be live, and a page cannot be opened.

In one embodiment, the preset problem information may be stored in a hot problem list, and when the electronic device determines a target recommended problem, the preset problem information is read from the hot problem list. And under the condition of successful reading, determining a target recommendation question according to the question intention classification information, the implementation acquired historical question information and the preset question information. And under the condition of reading failure, determining a target recommendation question according to the question intention classification information and the history question information obtained by implementing.

In the embodiment, the question of the user intention asking when the hot event occurs is predicted through the preset question information, and when the target recommendation question is determined by combining the preset question information, the relevance between the target recommendation question and the current situation encountered by the user can be improved, so that the accuracy of the target recommendation question is improved, the probability that the target recommendation question is selected as the question by the user is higher, the question intention of the user is accurately predicted, and the use experience of the user is improved. The target recommendation problem is obtained by combining the historical question information acquired in real time, the problem that the user lacks characteristic data during cold start is solved, the question intention of the user account can be preset under the condition that the newly registered user account lacks user portrait information, historical service information or service line identification corresponding to a historical operation page, the appropriate target recommendation problem is pushed to the user account, and the use experience of the user is improved. Furthermore, the historical question information is acquired in real time, so that the real-time performance of the historical question of the sample user account in the historical question information is high, the historical question information can accurately cover the question corresponding to the emergency event, the question of the user account can be quickly responded when the emergency event occurs on the platform, the target recommendation question can accurately cover the question which is intended to be provided by the user, the question which is not asked and answered first is achieved, the accuracy of the question recommended to the user is improved, and the use experience of the user is improved.

In one possible embodiment, the questioning intention classification information includes service line category information and question information corresponding to the service lines. In conjunction with FIG. 4, as shown in FIG. 5, S104b includes S104b1-S104b 4.

In S104b1, a first question to be recommended is determined according to a question in the question information whose confidence level is greater than a first preset threshold.

In one embodiment, the intent classification model is pre-trained by a Multi-task learning (MTL) structure. By training the intention classification model with a multi-task learning structure, the intention classification model can simultaneously execute two tasks. The first task is used for identifying service line information of user intention consultation, and the second task is used for identifying specific problems of user intention consultation.

In S104b2, the service line with the confidence level greater than the second preset threshold value in the service line type information is determined.

Alternatively, the confidence may be characterized by a value between 0 and 1.

In S104b3, the historical question questions belonging to the service line in the historical question information are determined as the second question to be recommended.

Optionally, the historical question questions belonging to the service line determined in S104b2 in the historical question information, for example, the historical question questions belonging to the live service line and the e-commerce service line, are obtained and determined as the second question to be recommended.

In S104b4, a target recommended question is determined according to the first question to be recommended, the second question to be recommended, the high-frequency question in the history question information, and the preset question information.

The high-frequency question is a question of which the question frequency in the historical question information meets a preset condition. For example, the question that satisfies the preset condition may be a question that the number of times the user actively asks satisfies a threshold. In one embodiment, the questions in the historical question information, of which the number of active questions meets a threshold value, are determined as high-frequency questions. For example, a question in the history question information, which is asked by the user for more than 100 times actively, may be determined as a high-frequency question. The threshold value may be adjusted according to practical situations, which is not limited by the present disclosure.

Optionally, when the electronic device obtains the first question to be recommended, the second question to be recommended, the high-frequency question in the historical question information, and the preset question information, the electronic device writes the first question to be recommended, the second question to be recommended, the high-frequency question in the historical question information, and the preset question information into the target result list. And then, the electronic equipment reads the problems meeting the preset conditions from the target result list and determines the problems as target recommendation problems.

In one embodiment, the questions in the target result list are sorted according to the order in which the questions obtained in different ways are written into the target result list, so as to obtain a target result queue. It can be understood that the first question to be recommended, the second question to be recommended, the high-frequency question in the historical question information and the preset question information are obtained through different ways.

In one embodiment, for a first to-be-recommended question, a second to-be-recommended question and preset question information, the accuracy of predicting the user question intention by the preset question information is the highest, the accuracy of predicting the user question intention by the first to-be-recommended question is lower than that of the preset question information, and the accuracy of predicting the user question intention by the second to-be-recommended question is lower than that of the first to-be-recommended question. Therefore, the first written target result list is the preset problem information, the second written target result list is the first to-be-recommended problem, and the third written target result list is the second to-be-recommended problem.

For example, when the terminal device displays the target recommendation question by turning pages, the question written into the target result list first may be displayed on the first page, and the question written into the target result list may be displayed on the subsequent page. For example, preset question information is displayed on a first page, a first question to be recommended is displayed on a second page, and a second question to be recommended is displayed on a third page.

It should be noted that when the number of the preset question information, the first question to be recommended, or the second question to be recommended is greater than the number that can be displayed on the page, the question that is ranked ahead is preferentially displayed. For example, for the preset problem information, the problem that the electronic device is ranked earlier when acquiring the preset problem information is preferentially displayed. And preferentially displaying the problem with high confidence in the first to-be-recommended problem aiming at the first to-be-recommended problem. And preferentially displaying the questions with more question times in the second questions to be recommended aiming at the second questions to be recommended.

In the embodiment, the question of the user intention asking when the hot event occurs is predicted through the preset question information, and when the target recommendation question is determined by combining the preset question information, the relevance between the target recommendation question and the current situation encountered by the user can be improved, so that the accuracy of the target recommendation question is improved, the probability that the target recommendation question is selected as the question by the user is higher, the question intention of the user is accurately predicted, and the use experience of the user is improved. And the question with the confidence level greater than the first preset threshold value in the question information is determined as the first question to be recommended, so that the first question to be recommended can more accurately cover the question intention of the user. Further, under the condition that a service line with the confidence coefficient larger than a second preset threshold exists in the service line category information, a problem corresponding to the service line information of the user intention problem is selected from the historical question information, so that a second problem to be recommended is obtained, and the second problem to be recommended can more accurately cover the problem of the user intention problem. Furthermore, the target recommendation problem obtained based on the first to-be-recommended problem, the second to-be-recommended problem and the preset problem information can be used for more accurately predicting the question-asking intention of the user, so that the accuracy of the target recommendation problem selected as the question by the user is improved, and the use experience of the user is further improved.

In one implementation possible embodiment, in a case that the questioning intention classification information includes service line classification information and question information corresponding to a service line, and there is no service line whose confidence is greater than a second preset threshold in the service line classification information, determining a target recommended question according to the questioning intention classification information, the historical questioning information, and the preset question information includes: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; and determining a target recommendation question according to the first to-be-recommended question, the high-frequency question in the historical question information and the preset question information.

In an embodiment, when there is no service line whose confidence is greater than the second preset threshold in the service line category information, that is, the confidence corresponding to the service line identifier in the service line category information is less than the second preset threshold, this indicates that the possibility that the service line identifier in the service line category information is the service line information that the user intends to consult is low. At the moment, the electronic equipment directly selects a high-frequency question from the historical question information, and determines a target recommendation question according to the first question to be recommended, the high-frequency question in the historical question information and preset question information.

In one possible implementation, the questioning intent classification information includes line of business category information. According to the questioning intention classification information, the historical questioning information and the preset question information, determining a target recommendation question, which comprises the following steps: determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining the historical question belonging to the service line in the historical question information as a second question to be recommended; determining a target recommendation problem according to the second to-be-recommended problem, the high-frequency problem in the historical question information and preset problem information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In the embodiment, the question of the user intention asking when the hot event occurs is predicted through the preset question information, and when the target recommendation question is determined by combining the preset question information, the relevance between the target recommendation question and the current situation encountered by the user can be improved, so that the accuracy of the target recommendation question is improved, the probability that the target recommendation question is selected as the question by the user is higher, the question intention of the user is accurately predicted, and the use experience of the user is improved. Further, under the condition that a service line with the confidence coefficient larger than a second preset threshold exists in the service line category information, a problem corresponding to the service line information of the user intention problem is selected from the historical question information, so that a second problem to be recommended is obtained, and the second problem to be recommended can more accurately cover the problem of the user intention problem. Furthermore, the question-asking intention of the user can be predicted more accurately based on the second question to be recommended and the target recommendation question obtained by the preset question information, so that the accuracy of the target recommendation question selected as the question by the user is improved, and the use experience of the user is improved.

In one possible embodiment, the questioning intent classification information includes question information corresponding to a line of business. According to the questioning intention classification information, the historical questioning information and the preset question information, determining a target recommendation question, which comprises the following steps: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining a target recommendation problem according to the first to-be-recommended problem, the high-frequency problem in the historical question information and preset problem information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In the embodiment, the question of the user intention asking when the hot event occurs is predicted through the preset question information, and when the target recommendation question is determined by combining the preset question information, the relevance between the target recommendation question and the current situation encountered by the user can be improved, so that the accuracy of the target recommendation question is improved, the probability that the target recommendation question is selected as the question by the user is higher, the question intention of the user is accurately predicted, and the use experience of the user is improved. And the question with the confidence level greater than the first preset threshold value in the question information is determined as the first question to be recommended, so that the first question to be recommended can more accurately cover the question intention of the user. Furthermore, by combining the first to-be-recommended question, the high-frequency question in the historical question information and the preset question information, the question which is asked by the user intention can be covered more accurately, so that the accuracy of predicting the question which is asked by the user is improved, the possibility that the target recommended question is selected as the question by the user is further improved, and the use experience of the user is improved.

In one possible implementation, as shown in fig. 6 in conjunction with fig. 1, S104 includes S104 c.

In S104c, a target recommended question is determined based on the questioning intention classification information and the preset question information.

And the preset problem information is determined according to the current hotspot event information.

Optionally, after obtaining the questioning intention classification information, the electronic device obtains the target recommended question according to the questioning intention information predicted by the questioning intention classification information and preset question information.

In one embodiment, the preset problem information may be stored in a hot problem list, and when the electronic device determines a target recommended problem, the preset problem information is read from the hot problem list. And under the condition of successful reading, determining a target recommendation question according to the question intention classification information and the preset question information. And under the condition of reading failure, determining a target recommendation question according to the questioning intention classification information.

In the embodiment, the question of the user intention asking when the hot event occurs is predicted through the preset question information, and when the target recommendation question is determined by combining the preset question information, the relevance between the target recommendation question and the current situation encountered by the user can be improved, so that the accuracy of the target recommendation question is improved, the probability that the target recommendation question is selected as the question by the user is higher, the question intention of the user is accurately predicted, and the use experience of the user is improved. Furthermore, the target recommendation question is obtained by combining the question intention classification information and the preset question information, and the question intention of the user can be predicted more accurately, so that the target recommendation question accurately covers the question which is supposed to be provided by the user, the question which is not asked and answered first is achieved, the accuracy of the question recommended to the user is improved, and the use experience of the user is improved.

In one possible embodiment, the questioning intent classification information includes question information corresponding to a line of business. According to the questioning intention classification information and the preset question information, determining a target recommendation question, which comprises the following steps: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; and determining a target recommendation problem according to the first to-be-recommended problem and preset problem information.

In the embodiment, the question of the user intention asking when the hot event occurs is predicted through the preset question information, and when the target recommendation question is determined by combining the preset question information, the relevance between the target recommendation question and the current situation encountered by the user can be improved, so that the accuracy of the target recommendation question is improved, the probability that the target recommendation question is selected as the question by the user is higher, the question intention of the user is accurately predicted, and the use experience of the user is improved. And the question with the confidence level greater than the first preset threshold value in the question information is determined as the first question to be recommended, so that the first question to be recommended can more accurately cover the question intention of the user. Furthermore, the target recommendation problem obtained by combining the first to-be-recommended problem and the preset problem information can be used for more accurately predicting the question-asking intention of the user, so that the accuracy of selecting the target recommendation problem as the question-asking problem by the user is improved, and the use experience of the user is improved.

In one possible implementation, as shown in FIG. 7, the question recommendation method further includes S105-S106.

In S105, user portrait information, historical service information, and a service line identifier corresponding to a historical operation page of each sample user account are obtained.

Optionally, the sample user account is a registered user account of the application.

Optionally, the user representation information for each sample user account is obtained from a local cache of the electronic device.

In one embodiment, the electronic device updates user profile information for user accounts in the application at a day granularity and stores the user profile information in a local cache for retrieval by the electronic device for each sample user account when training the intent classification model.

Alternatively, the historical service information of each sample user account on each service line can be obtained by reading the service line interface.

In one embodiment, the historical service information may be service operation information of the sample user account within a preset time period. For example, it may be business operation information for up to three months.

Optionally, the service line identifier corresponding to the sample user account historical operation page may be obtained by reading the service line interface in real time. The historical operation page is used for representing the pages which are historically browsed.

In an embodiment, if the sample user account browses the live broadcast page, the service line identifier corresponding to the historical operation page is the identifier of the live broadcast service line. And if the sample user account browses the e-commerce page, the service line identifier corresponding to the historical operation page is the identifier of the e-commerce service line. And if the sample user account browses the search page, the service line identifier corresponding to the historical operation page is the identifier of the search service line.

In S106, multitask training is carried out on the classification neural network model through the user portrait information, the historical business information and the business line identification corresponding to the historical operation page until the classification neural network model is converged, and an intention classification model is obtained.

The multi-task training comprises task training for identifying the class of a business line to be asked of a sample user account and task training for identifying the information of the question to be asked of the sample user account.

Optionally, the intention classification model is pre-trained by a Multi-task learning (MTL) structure. The intention classification model can simultaneously execute the first task and the second task by training the intention classification model through a multi-task learning structure. The first task is used for identifying service line information of user intention consultation, and the second task is used for identifying specific problems of user intention consultation.

Illustratively, the classification neural network model may be a Transformer model.

Illustratively, the line-of-business information that the user intends to consult may be one or more of live broadcast, e-commerce, commercialization, search, recommendation, security.

For example, the specific questions that the user intends to consult may be one or more of inquiry about logistics progress, inquiry about account status, refund, how to get popular.

In one embodiment, when the intention classification model is subjected to multi-task learning training, the underlying parameters (namely, portrait information, historical business information and business line identifications corresponding to historical operation pages) shared by the two tasks are influenced by the two tasks in parameter optimization. When the tasks are converged, the intention classification model equivalently fuses the two tasks, and the accuracy of each task is improved.

Illustratively, as shown in fig. 8, the feature list composed of the user portrait information, the historical service information, and the service line identifier corresponding to the historical operation page is located at the L1 level (i.e., the bottom layer). In the layer L1, the service line identifiers corresponding to the user portrait information, the historical service information, and the historical operation page respectively have corresponding neural network structures. The first task and the second task share a neural network structure of layer L1. At the L2 level, the first task and the second task correspond to a neural network structure. In the L3 layer, there are the service line category information output by the first task and the problem information corresponding to the service line output by the second task.

In one embodiment, at level L1, after the intent classification model obtains user image information, such as age, gender, … …, fan count, etc., the user image information is normalized to obtain user image characteristics U1, U2, … …, Un. U1, U2, … … and Un represent each of the user portrait information with a numerical value between 0 and 1, respectively.

In one embodiment, at the L1 level, after the intention classification model obtains the historical business information, for example, the historical business information may be an order, a report, … …, a live broadcast, and the like, each business operation feature in the historical business information is normalized to obtain historical business operation features S1, S2, … …, and Sn. Wherein, S1, S2, … …, Sn respectively use values between 0 and 1 to characterize each service operation information in the historical service information.

In one embodiment, at the level of L1, after the intention classification model obtains the service line identifiers corresponding to the historical operation pages, for example, the service line identifiers may be P1, P2, … …, Pm, each service line identifier is converted into a target vector through a coding model. Illustratively, the coding model may be an Encoder model. Optionally, the feature processing module (i.e. the first layer in fig. 8) of the intention classification model processes the original data of the user portrait information, the historical business information, and the business line identifier corresponding to the historical operation page, so as to obtain three types of feature data. Further, the three types of feature data are connected in series to serve as a model to be introduced into a prediction module of the intention classification model (namely, the second layer in fig. 8), and the service line information of the intention consultation of the sample user account and the specific problem of the intention consultation are identified and output by the prediction module of the intention classification model.

Optionally, the intent classification model may also be trained by single task learning, such that the intent classification model may perform the first task or the second task.

In the embodiment, the classification neural network model is subjected to multi-task training through the multi-dimensional features formed by fusing the user portrait information, the historical service information and the service line identifications corresponding to the historical operation pages to obtain the intention classification model, and all feature data which possibly cause customer complaints of the user are basically covered, so that the intention classification model can more accurately predict the questioning intention of the user. Further, through multi-task training, the intention classification model can simultaneously predict the service line information of the user intention consultation and the specific problems of the intention consultation, and the prediction accuracy is higher compared with a model which can only execute a single task.

The scheme provided by the embodiment of the application is mainly introduced from the perspective of a method. To implement the above functions, it includes hardware structures and/or software modules for performing the respective functions. Those of skill in the art would readily appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as hardware or combinations of hardware and computer software. Whether a function is performed as hardware or computer software drives hardware depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

The embodiment of the disclosure also provides a problem recommendation device.

FIG. 9 is a block diagram illustrating an issue recommendation device, according to an example embodiment. Referring to fig. 9, the question recommending apparatus 900 includes a determining module 901, an obtaining module 902, and a classifying module 903.

A determining module 901 configured to execute determining a user account identifier to be subjected to question recommendation. For example, in conjunction with fig. 1, the determining module 901 may be configured to perform S101.

The obtaining module 902 is configured to perform obtaining of user portrait information corresponding to a user account identifier, historical service information, and a service line identifier corresponding to a historical operation page. For example, in conjunction with fig. 1, the obtaining module 902 may be configured to perform S102.

And the classification module 903 is configured to execute the question intention classification information corresponding to the user account identification according to the user portrait information, the historical business information, the business line identification corresponding to the historical operation page and a pre-trained intention classification model. For example, in conjunction with fig. 1, the classification module 903 may be used to perform S103.

The determining module 901 is further configured to perform: and determining a target recommendation question according to the questioning intention classification information. For example, in conjunction with fig. 1, the determining module 901 may be configured to perform S104.

In one possible implementation, the determining module 901 is specifically configured to perform: determining a target recommendation question according to the question intention classification information and the historical question information; the historical question information includes historical question questions for a plurality of sample user accounts. For example, in conjunction with fig. 2, the determining module 901 may be used to execute S104 a.

In another possible implementation, the questioning intention classification information includes service line classification information and question information corresponding to a service line, and the determining module 901 is specifically configured to perform: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining the historical question belonging to the service line in the historical question information as a second question to be recommended; (ii) a Determining a target recommendation question according to the first question to be recommended, the second question to be recommended and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition. For example, in conjunction with FIG. 3, the determination module 901 may be used to perform S104a1-S104a 4.

In another possible implementation, the questioning intention classification information includes service line classification information, and the determining module 901 is specifically configured to perform: determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining a target question to be recommended according to a historical question belonging to a service line in the historical question information and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the questioning intention classification information includes question information corresponding to a service line, and the determining module 901 is specifically configured to perform: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining a target recommendation question according to the first question to be recommended and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the determining module 901 is specifically configured to perform: determining a target recommendation question according to the questioning intention classification information, the historical questioning information and the preset question information; the preset question information is determined according to the current hotspot event information, and the historical question information comprises historical question questions of a plurality of sample user accounts. For example, in conjunction with fig. 4, the determining module 901 may be used to execute S104 b.

In another possible implementation, the questioning intention classification information includes service line classification information and question information corresponding to the service line, and the determining module is specifically configured to perform: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining the historical question belonging to the service line in the historical question information as a second question to be recommended; determining a target recommendation problem according to the first to-be-recommended problem, the second to-be-recommended problem, a high-frequency problem in the historical question information and preset problem information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition. For example, in conjunction with FIG. 5, the determination module 901 may be used to perform S104b1-S104b 4.

In another possible implementation, the questioning intention classification information includes service line classification information, and the determining module 901 is specifically configured to perform: determining the service lines with the confidence level greater than a second preset threshold value in the service line category information; determining the historical question belonging to the service line in the historical question information as a second question to be recommended; determining a target recommendation problem according to the second to-be-recommended problem, the high-frequency problem in the historical question information and preset problem information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the questioning intention classification information includes question information corresponding to a service line, and the determining module 901 is specifically configured to perform: determining a first problem to be recommended according to the problem of which the confidence level in the problem information is greater than a first preset threshold value; determining a target recommendation problem according to the first to-be-recommended problem, the high-frequency problem in the historical question information and preset problem information; the high-frequency question is a question of which the question frequency in the historical question information meets the preset condition.

In another possible implementation, the determining module 901 is specifically configured to perform: determining a target recommendation question according to the questioning intention classification information and preset question information; and the preset problem information is determined according to the current hotspot event information. For example, in conjunction with fig. 6, the determining module 901 may be used to execute S104 c.

In another possible embodiment, the apparatus further includes a training module configured to perform: acquiring user portrait information, historical service information and a service line identifier corresponding to a historical operation page of each sample user account; performing multi-task training on the classification neural network model through user portrait information, historical service information and service line identification corresponding to a historical operation page until the classification neural network model is converged to obtain an intention classification model; the multi-task training comprises task training for identifying the class of a business line to be asked of a sample user account and task training for identifying the information of the question to be asked of the sample user account. For example, in connection with FIG. 7, a training module may be used to perform S105-S106.

With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.

FIG. 10 is a block diagram illustrating an electronic device in accordance with an example embodiment. As shown in fig. 10, the electronic device 1000 includes, but is not limited to: a processor 1001 and a memory 1002.

The memory 1002 is used for storing executable instructions of the processor 1001. It is understood that the processor 1001 is configured to execute instructions to implement the problem recommendation method shown in any one of fig. 1 to 8 in the above embodiments.

It should be noted that the electronic device structure shown in fig. 10 does not constitute a limitation of the electronic device, and the electronic device may include more or less components than those shown in fig. 10, or combine some components, or arrange different components, as will be understood by those skilled in the art.

The processor 1001 is a control center of the electronic device, connects various parts of the entire electronic device using various interfaces and lines, and performs various functions of the electronic device and processes data by running or executing software programs and/or modules stored in the memory 1002 and calling data stored in the memory 1002, thereby performing overall monitoring of the electronic device. Processor 1001 may include one or more processing units; optionally, the processor 1001 may integrate an application processor and a modem processor, wherein the application processor mainly handles operating systems, user interfaces, application programs, and the like, and the modem processor mainly handles wireless communications. It will be appreciated that the modem processor described above may not be integrated into the processor 1001.

The memory 1002 may be used to store software programs as well as various data. The memory 1002 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program (such as an image acquisition module, a positioning module, a hidden variable acquisition module or an image generation module) required by at least one functional module, and the like. Further, the memory 1002 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device.

In an exemplary embodiment, the disclosed embodiment also provides a computer-readable storage medium comprising instructions, such as the memory 1002 comprising instructions, which are executable by the processor 1001 of the electronic device 1000 to perform the problem recommendation method shown in any one of fig. 1 to 7 described above.

In practical implementation, the processing functions of the determining module 901, the obtaining module 902 and the classifying module 903 can be realized by the processor 1001 shown in fig. 10 calling the program code in the memory 1002. The specific implementation process may refer to the description of the problem recommendation method portion shown in any one of fig. 1 to 7, and is not described herein again.

Alternatively, the computer-readable storage medium may be a non-transitory computer-readable storage medium, which may be, for example, a Read-Only Memory (ROM), a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

In an exemplary embodiment, the disclosed embodiments also provide a computer program product comprising one or more instructions executable by the processor 1001 of the electronic device 1000 to perform the problem recommendation method shown in any one of fig. 1 to 7 described above.

It should be noted that the instructions in the computer-readable storage medium or one or more instructions in the computer program product are executed by the processor 1001 of the electronic device 1000 to implement the processes of the problem recommendation method embodiment, and can achieve the same technical effect as the problem recommendation method shown in any one of fig. 1 to 7, and in order to avoid repetition, the description is not repeated here.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A question recommendation method, comprising:

determining a user account identifier to be subjected to problem recommendation;

acquiring user portrait information, historical service operation information and a service line identifier corresponding to a historical operation page, wherein the user portrait information and the historical service operation information correspond to the user account identifier;

according to the user portrait information, historical business operation information, business line identifications corresponding to historical operation pages and a pre-trained intention classification model, determining question intention classification information corresponding to the user account identifications;

and determining a target recommendation question according to the questioning intention classification information.

2. The question recommendation method according to claim 1, wherein determining a target recommendation question according to the questioning intention classification information comprises:

determining a target recommendation question according to the question intention classification information and the historical question information; the historical question information includes historical question questions for a plurality of sample user accounts.

3. The question recommendation method according to claim 2, wherein the question intention classification information includes service line classification information and question information corresponding to service lines, and the determining a target recommendation question according to the question intention classification information and historical question information includes:

determining a first to-be-recommended problem according to the problem of which the confidence level in the problem information is greater than a first preset threshold;

determining the service line with the confidence level greater than a second preset threshold value in the service line category information;

determining the historical question belonging to the service line in the historical question information as a second question to be recommended;

determining a target recommendation question according to the first question to be recommended, the second question to be recommended and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets a preset condition.

4. The question recommendation method according to claim 2, wherein the question intention classification information includes service line classification information, and the determining a target recommendation question according to the question intention classification information and the historical question information includes:

determining a target question to be recommended according to a historical question belonging to the service line in the historical question information and a high-frequency question in the historical question information; the high-frequency question is a question of which the question frequency in the historical question information meets a preset condition.

5. The question recommendation method according to claim 1, wherein determining a target recommendation question according to the questioning intention classification information comprises:

determining a target recommendation question according to the questioning intention classification information, the historical questioning information and preset question information; the preset question information is determined according to the current hotspot event information, and the historical question information comprises historical question questions of a plurality of sample user accounts.

6. The question recommendation method according to claim 1, wherein determining a target recommendation question according to the questioning intention classification information comprises:

determining a target recommendation question according to the questioning intention classification information and preset question information; and the preset problem information is determined according to the current hotspot event information.

7. The question recommendation method according to any one of claims 1-6, characterized in that the intention classification model is pre-trained by:

acquiring user portrait information, historical service operation information and a service line identifier corresponding to a historical operation page of each sample user account;

performing multi-task training on a classification neural network model through the user portrait information, the historical service operation information and the service line identification corresponding to the historical operation page until the classification neural network model is converged to obtain the intention classification model; the multi-task training comprises task training for identifying the class of a business line to be asked of a sample user account and task training for identifying the information of the question to be asked of the sample user account.

8. A question recommendation device, comprising:

the determining module is configured to execute the step of determining the user account identifier to be subjected to question recommendation;

the acquisition module is configured to execute the acquisition of user portrait information, historical service information and service line identification corresponding to a historical operation page corresponding to the user account identification;

the classification module is configured to execute the question intention classification information corresponding to the user account identification according to the user portrait information, the historical service information, the service line identification corresponding to the historical operation page and a pre-trained intention classification model;

the determination module is further configured to perform: and determining a target recommendation question according to the questioning intention classification information.

9. An electronic device, comprising:

a processor;

a memory for storing the processor-executable instructions;

wherein the processor is configured to execute the instructions to implement the issue recommendation method of any one of claims 1 to 7.

10. A computer-readable storage medium, wherein instructions in the computer-readable storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the issue recommendation method of any of claims 1-7.