CN-116467526-B - Multi-task recommendation method and device integrating priori information

CN116467526BCN 116467526 BCN116467526 BCN 116467526BCN-116467526-B

Abstract

The application provides a multitask recommendation method and device integrating priori information, which relate to the field of artificial intelligence and can be also used in the financial field, and the method and device comprise the steps of generating expert characterization vectors corresponding to all experts and user characterization vectors corresponding to current questioning users according to buried data corresponding to the questioning and answering priori information; the embedded point data comprises full-quantity user questioning data and full-quantity expert answering data, the questioning and answering priori information comprises current questioning information of the current questioning user, a plurality of experts corresponding to the current questioning information are determined according to the user characterization vector and the expert characterization vector, and optimal answers corresponding to the current questioning information are recommended from all answers provided by the plurality of experts. The application can fuse the prestored question and answer priori information to carry out multitask recommendation, and help users to quickly search the optimal answer of the questions presented by the users by utilizing the strong correlation among the data.

Inventors

DENG WEI
REN YUPING
SHI ZHONGDE
WANG XIAOHONG

Assignees

中国工商银行股份有限公司

Dates

Publication Date: 20260505
Application Date: 20230425

Claims (8)

1. A multitask recommendation method fusing priori information, comprising: when a current questioning user issues current questioning information, adding the current questioning information and the current questioning user into the questioning and answering priori information; Inputting buried data corresponding to the question-answering priori information into an artificial neural network to obtain a full-quantity question word vector and a full-quantity answer word vector, wherein the buried data comprises full-quantity user question data and full-quantity expert answer data; Constructing a user characterization vector based on the question vector corresponding to the current question user in the full-quantity question vector, wherein the user characterization vector is used for characterizing all questions which the user has posed historically; Constructing an expert characterization vector based on the answer word vector corresponding to each expert in the full-quantity return answering speech vector, wherein the expert characterization vector is used for characterizing all questions answered by the expert in history; Respectively calculating the similarity between expert characterization vectors corresponding to the experts and the user characterization vectors to obtain a first similarity value; Performing normalization processing and sequencing processing on the first similarity value to obtain a plurality of experts corresponding to the current question information; selecting answer word vectors corresponding to all answers provided by the plurality of experts from the full-quantity return answering speech vectors; respectively calculating the similarity between each answer word vector and the question word vector corresponding to the current question user to obtain a second similarity value; and selecting the answer corresponding to the largest one of the second similarity values as the optimal answer.
2. The method of claim 1, further comprising, after recommending an optimal answer corresponding to the current question information from all answers provided by the plurality of experts: constructing a recommended error loss function according to the first similarity value and the second similarity value; and adjusting parameters of the artificial neural network according to the recommended error loss function.
3. The method for multitasking recommendation with fusion of a priori information according to claim 2, wherein constructing a recommendation error loss function according to the normalized first similarity value and the second similarity value comprises: calculating an expert recommendation error loss function according to the first similarity value; calculating an answer recommendation error loss function according to the second similarity value; And constructing a recommended error loss function according to the expert recommended error loss function and the answer recommended error loss function.
4. A method of multitasking for a fusion of a priori information according to claim 3, in which said adjusting parameters of said artificial neural network according to said recommended error loss function comprises: calculating a loss value corresponding to the artificial neural network by using the recommended error loss function; And adjusting parameters of the artificial neural network according to the loss value until the recommended error loss function reaches a convergence state.
5. A multitasking recommendation device incorporating a priori information, comprising: The system comprises a characteristic vector generation unit, a query vector input unit, a query vector generation unit and a query vector generation unit, wherein the current query information and the current query user are added to query priori information when the current query user issues the current query information, the query vector generation unit is used for inputting buried data corresponding to the query priori information into an artificial neural network to obtain a full-scale query word vector and a full-scale answer word vector, the buried data comprise full-scale user query data and full-scale expert answer data, the query priori information comprises the current query information of the current query user, a user characteristic vector is constructed based on the query word vector corresponding to the current query user in the full-scale query word vector, the user characteristic vector is used for representing all questions which the user has submitted in history, the expert characteristic vector is used for representing all questions which the expert has answered in history based on the full-scale answer answering speech vector; the expert recommendation unit is used for respectively calculating the similarity between the expert characterization vector corresponding to each expert and the user characterization vector to obtain a first similarity value, and carrying out normalization processing and sorting processing on the first similarity value to obtain a plurality of experts corresponding to the current question information; The answer recommendation unit is used for selecting answer word vectors corresponding to all answers provided by the plurality of experts from the full-quantity return answering speech vectors, respectively calculating the similarity between each answer word vector and the question word vector corresponding to the current question user to obtain a second similarity value, and selecting the answer corresponding to the largest one of the second similarity values as the optimal answer.
6. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the multitasking recommendation method of fusing a priori information of any of claims 1 to 4 when the program is executed by the processor.
7. A computer readable storage medium having stored thereon a computer program, which when executed by a processor performs the steps of the method of multitasking recommendation incorporating a priori information as claimed in any of claims 1 to 4.
8. A computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the method for multitasking recommendation incorporating a priori information of any of claims 1 to 4.

Description

Multi-task recommendation method and device integrating priori information Technical Field The application relates to the field of artificial intelligence, and can be used in the financial field, in particular to a multi-task recommendation method and device integrating priori information. Background Information retrieval (Information Retrieval) is one way for a user to perform information retrieval. The narrow information search mainly refers to information query (Information Search), i.e. a process in which a user inputs his or her question according to the need and, by means of a search tool, finds out the required information from the information collection. Based on the above, the financial field introduces an online communication platform for popular questions and answers. Information retrieval is one of the main functions. How to ensure that a user can quickly and efficiently acquire ideal answers in the process of information retrieval is an important way for improving the viscosity of the user and accumulating good public praise for products. In practical application, the problem solving rate and the solution timeliness are important user experience indexes. How to address the questions of a user is a technical difficulty in the existing expert knowledge base, and at the same time, retrieving the expert and answer for the user that are most likely to solve the questions. Disclosure of Invention Aiming at the problems in the prior art, the application provides a multi-task recommendation method and device integrating priori information, which can integrate prestored question and answer priori information to carry out multi-task recommendation and help users to quickly search the optimal answers of the questions presented by the users by utilizing the strong correlation among data. In order to solve the technical problems, the application provides the following technical scheme: In a first aspect, the present application provides a method for multitasking recommendation with fusion of priori information, including: Generating expert characterization vectors corresponding to all experts and user characterization vectors corresponding to current questioning users according to buried data corresponding to the questioning and answering priori information, wherein the buried data comprises full-quantity user questioning data and full-quantity expert answering data; determining a plurality of experts corresponding to the current question information according to the user characterization vector and the expert characterization vector; and recommending the optimal answer corresponding to the current question information from all answers provided by the plurality of experts. Further, the generating, according to the buried data corresponding to the question-answer priori information, an expert characterization vector corresponding to each expert and a user characterization vector corresponding to the current question user includes: Inputting the full-quantity user question data and the full-quantity expert answer data into a long-short-term memory artificial neural network to obtain a full-quantity question word vector and a full-quantity answer word vector; Constructing the user characterization vector based on the question word vector corresponding to the current question user in the full-quantity question word vector; and constructing the expert characterization vector based on the answer word vector corresponding to each expert in the full-quantity return answering speech vector. Further, the determining, according to the user characterization vector and the expert characterization vector, a plurality of experts corresponding to the current question information includes: Respectively calculating the similarity between expert characterization vectors corresponding to the experts and the user characterization vectors to obtain a first similarity value; and carrying out normalization processing and sequencing processing on the first similarity value to obtain a plurality of experts corresponding to the current question information. Further, the recommending the optimal answer corresponding to the current question information from all answers provided by the plurality of experts includes: selecting answer word vectors corresponding to all answers provided by the plurality of experts from the full-quantity return answering speech vectors; respectively calculating the similarity between each answer word vector and the question word vector corresponding to the current question user to obtain a second similarity value; and selecting the answer corresponding to the largest one of the second similarity values as the optimal answer. Further, after recommending the optimal answer corresponding to the current question information from all answers provided by the plurality of experts, the method further comprises: constructing a recommended error loss function according to the first similarity value and the second simi