CN-121981247-A - Question-answering processing method, question-answering processing device, electronic equipment, medium and program product

CN121981247ACN 121981247 ACN121981247 ACN 121981247ACN-121981247-A

Abstract

The disclosure provides a question-answering processing method, a question-answering processing device, electronic equipment, a medium and a program product, relates to application of a large model in intelligent customer service, and can be applied to the technical field of artificial intelligence, the technical field of big data and the technical field of finance and technology. The method comprises the steps of obtaining problem text information submitted by a user, determining corresponding semantic attributes based on the problem text, performing retrieval in a knowledge vector base based on the problem text information and the semantic attributes to obtain target knowledge content, constructing model input content based on the problem text information, the semantic attributes and the target knowledge content, inputting the model input content into a question and answer processing model to obtain answer information corresponding to the problem text information, generating answer content based on the answer information, and returning the answer content to the user.

Inventors

LIANG LINA
SHI ZHONGDE
YANG KE
YANG KAI

Assignees

中国工商银行股份有限公司

Dates

Publication Date: 20260505
Application Date: 20250701

Claims (13)

1. A question-answering processing method, characterized in that the method comprises: acquiring problem text information submitted by a user, and determining corresponding semantic attributes based on the problem text information; searching is carried out in a knowledge vector base based on the problem text information and the semantic attribute, and target knowledge content is obtained; Building model input content based on the question text information, the semantic attribute and the target knowledge content, inputting the model input content into a question-answer processing model to obtain answer information corresponding to the question text information, and And generating response content based on the response information, and returning the response content to the user.
2. The method of claim 1, wherein the determining the corresponding semantic attributes based on the question text information comprises: word segmentation processing is carried out on the problem text information, and a candidate word set is obtained; based on the candidate word set, word frequency statistics is carried out in a preset corpus, and a characteristic value is calculated based on a statistical result; Sorting the feature values, selecting m items as problem characterization vectors based on the sorting result, wherein m is a positive integer, and Inputting the problem representation vector into a semantic classification model to obtain the semantic attribute, wherein the semantic attribute at least comprises a problem type label and a first service field label corresponding to the problem text information.
3. The method according to claim 2, wherein the method further comprises: calculating the confidence coefficient of the first service field label; In response to the confidence level being less than a first target threshold, performing semantic similarity calculation on the candidate word set and domain description texts of a plurality of preset service domains, determining a second service domain label with a similarity value greater than a second target threshold, and And combining the first service domain label and the second service domain label to form composite service domain information, and determining the semantic attribute based on the composite service domain information.
4. The method of claim 1, wherein the constructing of the knowledge vector base comprises: Carrying out structural transformation on the knowledge document to obtain a structural document; executing regular expression extraction based on a custom field rule on the structured document to obtain a structured text; paragraph classification of the structured text based on a linguistic analysis model, and And constructing a plurality of question-answer pairs based on the classification result, converting the question-answer pairs into vector representations, and writing the vector representations into the knowledge vector base.
5. A method according to claim 3, wherein prior to entering the model input content into a question-answer process model, the method further comprises: acquiring a question type prompt word corresponding to the question type label and And integrating the question type prompt word with the corresponding question text information.
6. The method of claim 5, wherein prior to inputting the model input content into a question-answer process model, the method further comprises: acquiring an expert role template corresponding to the first service domain label or the composite service domain information; acquiring role identity prompt words based on the expert role template and And integrating the character identity prompt words with the corresponding problem text information.
7. The method according to any one of claims 1-6, wherein the performing the search in a knowledge vector base based on the question text information and the semantic attribute to obtain a target knowledge content includes: Selecting a target retrieval strategy based on the question text information; Performing preliminary search in the knowledge vector base based on the target search strategy, performing context segment screening based on the preliminary search result to obtain correlation content, and And filtering and extracting the correlation content to obtain the target knowledge content.
8. The method of any one of claims 1-6, wherein the constructing model input content based on the question text information, the semantic attributes, and the target knowledge content comprises: Generating a context prompt block based on the target knowledge content, integrating the context prompt block with the question text information and the semantic attributes to form an enhanced input sequence, and And acquiring an industry standard constraint prompt word, and integrating the industry standard constraint prompt word in the enhanced input sequence to construct the model input content.
9. The method according to claim 4, wherein the method further comprises: Acquiring feedback information of the user on the response content; responding to the feedback information to be effective, and calculating semantic similarity between the problem text information and each problem in the knowledge vector library; if the semantic similarity is smaller than a third target threshold, forming a supplementary question-answer pair by the question text information and the answer content, and Writing the supplementary question-answer pairs into the knowledge vector base.
10. A question-answering apparatus, the apparatus comprising: The data acquisition module is used for acquiring problem text information submitted by a user and determining corresponding semantic attributes based on the problem text information; the knowledge base searching module is used for performing searching in a knowledge vector base based on the problem text information and the semantic attribute to obtain target knowledge content; A question-answer processing module for constructing model input content based on the question text information, the semantic attribute and the target knowledge content, inputting the model input content into a question-answer processing model to obtain answer information corresponding to the question text information, and And the response generation module is used for generating response content based on the response information and returning the response content to the user.
11. An electronic device, comprising: One or more processors; a memory for storing one or more computer programs, Characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1-9.
12. A computer-readable storage medium, on which a computer program or instructions is stored, which, when executed by a processor, carries out the steps of the method according to any one of claims 1 to 9.
13. A computer program product comprising a computer program or instructions which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 9.

Description

Question-answering processing method, question-answering processing device, electronic equipment, medium and program product Technical Field The present disclosure relates to application of a large model in intelligent customer service, and more particularly to an artificial intelligence technology field, a large data technology field, and a financial science field, and more particularly to a question-answering processing method, apparatus, device, medium, and program product. Background Question and answer applications in the financial field usually adopt a keyword matching or rule template-based mode to perform question understanding and answer retrieval, but the mode has more limitations in practical application. On the one hand, the existing question-answering system lacks the capability of precisely identifying the user questions, and cannot accurately judge the types of the questions and the financial service field, so that the answers provided by the system cannot meet the user requirements frequently, and the situation of answering questions is presented. On the other hand, the existing knowledge base has a single structure, mainly exists in a flattened form, and lacks of structural management of different types of financial knowledge such as system clauses, business processes, definition terms and the like. In addition, although some systems start to try to introduce a large language model to promote the intelligence of answer generation, due to lack of close combination with domain knowledge, the generated content often has problems such as generalization, fuzzy expression or non-compliance with industry specifications, and the like, so that the professionality and compliance of the answer content are difficult to guarantee. Disclosure of Invention In view of the foregoing, the present disclosure provides a question-answering processing method, apparatus, device, medium, and program product. According to a first aspect of the disclosure, a question-answer processing method is provided, and the method comprises the steps of obtaining question text information submitted by a user, determining corresponding semantic attributes based on the question text, performing retrieval in a knowledge vector base based on the question text information and the semantic attributes to obtain target knowledge content, constructing model input content based on the question text information, the semantic attributes and the target knowledge content, inputting the model input content into a question-answer processing model to obtain answer information corresponding to the question text information, generating answer content based on the answer information, and returning the answer content to the user. According to the embodiment of the disclosure, the method for determining the corresponding semantic attributes based on the question text comprises the steps of carrying out word segmentation processing on the question text information to obtain a candidate word set, carrying out word frequency statistics in a preset corpus based on the candidate word set, calculating feature values based on statistical results, sorting the feature values, selecting m items based on sorting results as question characterization vectors, wherein m is a positive integer, inputting the question characterization vectors into a semantic classification model to obtain the semantic attributes, wherein the semantic attributes at least comprise question type labels and first service field labels corresponding to the question text information. According to the embodiment of the disclosure, the method further comprises the steps of calculating the confidence coefficient of the first service domain label, responding to the fact that the confidence coefficient is smaller than a first target threshold value, conducting semantic similarity calculation on the candidate word set and domain description texts of a plurality of preset service domains, determining a second service domain label with a similarity value larger than a second target threshold value, combining the first service domain label and the second service domain label to form composite service domain information, and determining the semantic attribute based on the composite service domain information. According to the embodiment of the disclosure, the knowledge vector base is constructed by carrying out structural transformation on a knowledge document to obtain a structured document, carrying out regular expression extraction on the structured document based on a user-defined field rule to obtain a structured text, carrying out paragraph classification on the structured text based on a language analysis model, constructing a plurality of question-answer pairs based on the classification result, converting the question-answer pairs into vector representations, and writing the vector representations into the knowledge vector base. According to the embodiment of the disclosure, before the mode