CN-121996751-A - Intelligent question-answering method, device, equipment and medium

CN121996751ACN 121996751 ACN121996751 ACN 121996751ACN-121996751-A

Abstract

The invention relates to the technical field of artificial intelligence and discloses an intelligent question-answering method, device, equipment and medium, which comprise the steps of analyzing the similarity between a plurality of semantic vectors and vectors corresponding to user problems, and screening multi-source corpus data according to the similarity; extracting entity and semantic association relation in target corpus data, constructing an initial association network, dynamically updating the initial association network, matching user questions with the entity in the dynamic knowledge graph, screening initial candidate intentions corresponding to the user questions, identifying target candidate intentions corresponding to grammar entities of the user questions, determining target intentions corresponding to the user questions, identifying potential intentions corresponding to the user questions according to the initial candidate intentions, determining question-answering intentions corresponding to the user questions based on the potential intentions and the target intentions, and inquiring the question-answering intentions to obtain question-answering results of the user questions. The invention can improve the accuracy of intelligent question answering of the user.

Inventors

LIU YUNSHU
ZENG CHEN
DU JUAN

Assignees

招商局金融科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251218

Claims (10)

1. An intelligent question-answering method is characterized by comprising the following steps: semantic vectorization is carried out on multisource corpus data acquired in advance, so that a plurality of semantic vectors are obtained; Analyzing the similarity between the plurality of semantic vectors and the vector corresponding to the pre-acquired user problem, screening out a plurality of semantic vectors with the similarity meeting the preset screening condition, and determining corpus data corresponding to each screened semantic vector in the multi-source corpus data as target corpus data; Performing entity identification and relation extraction on the target corpus data to obtain a preset type entity and a semantic association relation between the preset type entity, constructing an initial association network according to the preset type entity and the semantic association relation, and dynamically updating the initial association network based on a preset updating strategy to obtain a dynamic knowledge graph; Calculating the association degree between the entity in the user problem and the business intention entity in the dynamic knowledge graph, and screening one or more business intention entities with association degree larger than a preset association threshold as initial candidate intents corresponding to the user problem; Extracting grammar entity corresponding to main guest in the user problem, matching the grammar entity with business intention entity corresponding to the initial candidate intention to obtain a matching degree value, and selecting the business intention entity with the largest matching degree value as a target intention; Identifying potential intentions corresponding to the user problems according to the context logic of the initial candidate intentions in the dynamic knowledge graph, and fusing the potential intentions and the target intentions into question-answering intentions corresponding to the user problems; And inquiring information matched with the question-answer intentions in a plurality of preset knowledge bases to obtain answers to the user questions.
2. The intelligent question-answering method according to claim 1, wherein the semantic vectorization of the pre-collected multi-source corpus data to obtain a plurality of semantic vectors includes: Identifying the file type of the multi-source corpus data, and carrying out field splitting on the multi-source corpus data based on the file type to obtain field content; performing word segmentation processing on the multi-source corpus data according to the field content to obtain word segmentation words corresponding to the multi-source corpus data; Updating the word segmentation words based on a preset updating strategy, and converting the updated word segmentation words into semantic vectors one by one.
3. The intelligent question-answering method according to claim 1, wherein the constructing an initial association network according to the preset type of entity and the semantic association relationship comprises: Identifying the entity type of the entity and identifying the relation type of the semantic association relation; Associating entities with the same entity type and relationship type with semantic association relationship to obtain an association path, and constructing a target network of the entity with the same type and semantic association relationship according to the association path; and identifying path association nodes in the semantic association relationship, and connecting the target network into an initial association network according to the path association nodes.
4. The intelligent question-answering method according to claim 1, wherein the dynamically updating the initial association network based on a preset updating policy to obtain a dynamic knowledge graph comprises: Extracting a timing update strategy and a trigger update strategy in the update strategy; scheduling a timing task based on the timing update strategy, and synchronously updating entity attributes in the initial association network according to the positioning task; Based on the trigger update strategy, monitoring text content change events corresponding to the entity and the semantic association relationship in the initial association network, and triggering a real-time update task of the initial association network when the monitored text content change events have the association relationship with the entity and the semantic association relationship in the initial association network; Updating the map relation in the initial association network according to the real-time updating task; and generating a dynamic knowledge graph according to the updated entity attributes and graph relationships.
5. The intelligent question-answering method according to claim 1, wherein the extracting the grammar entity corresponding to the main predicate in the user question comprises: Carrying out syntactic analysis on the user problem to obtain a main guest structure in the user problem; identifying a first entity corresponding to a subject, a second entity corresponding to a predicate and a third entity corresponding to an object in the subject-to-object structure; And taking the first entity, the second entity and the third entity as grammar entities in the user problem.
6. The intelligent question-answering method according to claim 1, wherein the identifying the potential intent corresponding to the user question according to the context logic of the initial candidate intent in the dynamic knowledge-graph comprises: querying entity data associated with the initial candidate intention in the dynamic knowledge graph; analyzing the association relationship type between the entity data and the initial candidate intention; extracting supplementary information corresponding to the user problem from the dynamic knowledge graph according to the association relationship type and the context logic; and combining the supplemental information with the intention answers corresponding to the initial candidate intention to obtain the potential intention corresponding to the user problem.
7. The intelligent question-answering method according to claim 1, wherein the querying information matched with the question-answering intention in a plurality of preset knowledge bases to obtain an answer to the user question comprises: determining query routing weights of the plurality of knowledge bases based on the question-answer intents and preset user attributes; performing search operation on the question and answer intents in parallel in the multiple knowledge bases according to the query routing weight to obtain a search result set; Calculating the similarity between each search result in the search result set and the problem intention, and primarily sequencing each search result in the search result set according to the sequence of the similarity from large to small; Screening out a plurality of search results of which the search results after preliminary sequencing meet the preset level label conditions; and sequencing the plurality of search results by using a preset sequencing algorithm, and integrating the sequenced search results into answers of the user questions.
8. An intelligent question-answering device, comprising: The semantic vectorization module is used for carrying out semantic vectorization on the multisource corpus data acquired in advance to obtain a plurality of semantic vectors; The multi-source corpus data screening module is used for analyzing the similarity between the plurality of semantic vectors and the vector corresponding to the pre-acquired user problem, screening out the plurality of semantic vectors with the similarity meeting the preset screening condition, and determining corpus data corresponding to each screened semantic vector in the multi-source corpus data as target corpus data; The dynamic knowledge graph construction module is used for carrying out entity identification and relation extraction on the target corpus data to obtain a preset type entity and a semantic association relation between the preset type entity, constructing an initial association network according to the preset type entity and the semantic association relation, and carrying out dynamic update on the initial association network based on a preset update strategy to obtain a dynamic knowledge graph; The initial candidate intention screening module is used for calculating the association degree between the entity in the user problem and the business intention entity in the dynamic knowledge graph, and screening one or more business intention entities with the association degree larger than a preset association threshold value as initial candidate intents corresponding to the user problem; The target intention determining module is used for extracting grammar entities corresponding to main guests in the user problem, matching the grammar entities with business intention entities corresponding to the initial candidate intention to obtain a matching degree value, and selecting the business intention entity with the largest matching degree value as a target intention; The question and answer intention determining module is used for logically identifying potential intentions corresponding to the user problems according to the context of the initial candidate intentions in the dynamic knowledge graph, and fusing the potential intentions and the target intentions into question and answer intentions corresponding to the user problems; And the question and answer result generation module is used for inquiring information matched with the question and answer intention in a plurality of preset knowledge bases to obtain the answer of the user question.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the intelligent question-answering method according to any one of claims 1 to 7 when the computer program is executed by the processor.
10. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the intelligent question-answering method according to any one of claims 1 to 7.

Description

Intelligent question-answering method, device, equipment and medium Technical Field The invention relates to the technical field of artificial intelligence, in particular to an intelligent question-answering method, device, equipment and medium. Background Along with the digitized transformation of enterprises entering into deep water areas, the deep energization of artificial intelligence technology in the field of system management becomes a key path for improving the efficiency of compliance management. In the group type enterprise management practice, the system of system intelligent question-answering is used as a core carrier of the compliance management, and still faces multiple technical bottlenecks. Most of the existing intelligent question-answering techniques are question-answering systems based on natural language large models, cannot distinguish the system difference among group departments, are difficult to respond to hierarchical requirements, and are difficult to match complex requirements of group enterprises under the background of industrial chain compliance upgrading and business diversification, so that the accuracy rate of intelligent question-answering to users is low. Disclosure of Invention The invention provides an intelligent question-answering method, device, equipment and medium, which are used for solving the technical problem of low accuracy rate when a user performs intelligent question-answering. In a first aspect, an intelligent question-answering method is provided, including: semantic vectorization is carried out on multisource corpus data acquired in advance, so that a plurality of semantic vectors are obtained; Analyzing the similarity between the plurality of semantic vectors and the vector corresponding to the pre-acquired user problem, screening out a plurality of semantic vectors with the similarity meeting the preset screening condition, and determining corpus data corresponding to each screened semantic vector in the multi-source corpus data as target corpus data; Performing entity identification and relation extraction on the target corpus data to obtain a preset type entity and a semantic association relation between the preset type entity, constructing an initial association network according to the preset type entity and the semantic association relation, and dynamically updating the initial association network based on a preset updating strategy to obtain a dynamic knowledge graph; Calculating the association degree between the entity in the user problem and the business intention entity in the dynamic knowledge graph, and screening out one or more business intention entities with the association degree larger than a preset association threshold as initial candidate intentions corresponding to the user problem; Extracting grammar entity corresponding to main guest in the user problem, matching the grammar entity with business intention entity corresponding to the initial candidate intention to obtain a matching degree value, and selecting the business intention entity with the largest matching degree value as a target intention; Identifying potential intentions corresponding to the user problems according to the context logic of the initial candidate intentions in the dynamic knowledge graph, and fusing the potential intentions and the target intentions into question-answering intentions corresponding to the user problems; And inquiring information matched with the question-answer intentions in a plurality of preset knowledge bases to obtain answers to the user questions. In a second aspect, an intelligent question answering device is provided, including: The semantic vectorization module is used for carrying out semantic vectorization on the multisource corpus data acquired in advance to obtain a plurality of semantic vectors; the multi-source corpus data screening module is used for analyzing the similarity between the plurality of semantic vectors and the vectors corresponding to the pre-acquired user problems, screening semantic vectors with the similarity meeting preset screening conditions, and determining corpus data corresponding to each screened semantic vector in the multi-source corpus data as target corpus data; The dynamic knowledge graph construction module is used for carrying out entity identification and relation extraction on the target corpus data to obtain a preset type entity and a semantic association relation between the preset type entity, constructing an initial association network according to the preset type entity and the semantic association relation, and carrying out dynamic update on the initial association network based on a preset update strategy to obtain a dynamic knowledge graph; The initial candidate intention screening module is used for calculating the association degree between the entity in the user problem and the business intention entity in the dynamic knowledge graph, and screening one or more business intention en