Industry-oriented context omission question-answering method

Abstract

The invention discloses an industry-oriented context omission question-answering method. According to the method, an industry knowledge graph is constructed according to structured and unstructured knowledge in industry data, the industry knowledge graph comprises triad information, and when a system determines a certain theme through a client problem, the theme and attributes related to the client problem in a triad are both identified as a dialogue state; and based on the industry knowledge graph and the determined dialogue state, the system performs dialogue state management after receivingthe client question, including solving noise interference through a historical state question and answer mechanism and a context omission question and answer judgment mechanism, emptying the system state caused by non-industry question and answer, and answering non-asked questions.

Description

technical field [0001] The invention relates to an industry-oriented question answering method for context omission, which belongs to the fields of natural language processing and machine learning. Background technique [0002] Dialogue State Management (Dialogue State Management) takes user intent as input, manages the dialog state generated during the contextual interaction process, and queries the knowledge base to generate a response. Natural, smooth, and efficient human-computer multi-round interaction experience is inseparable from efficient dialogue state management, but there are still many challenges in the actual engineering practice. First of all, the dialogue state is often determined by means of slot values ​​and entities, which will result in a large labeling cost, which will consume a large cost of manpower and material resources; in addition, the man-machine dialogue in the hall and other scenes is often mixed with a lot of noise (invalid question), when the...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the above-mentioned technical problems, the technical solution adopted in the present invention is: an industry-oriented context-eliminated question-and-answer method. First, an industry knowledge graph is constructed based on structured and unstructured knowledge in industry data. The industry knowledge graph includes three Tuple information, when the system determines a topic through customer questions, the topics and attributes related to the customer question in the triplet are identified as the dialogue state; based on the industry knowledge map and the determined dialogue state, after the system receives the customer question Dialogue state management, including solving noise interference through the historical state question-and-answer mechanism, context-ellipsis question-and-answer judgment mechanism, clearing the system state caused by non-industry questions and answers, and answering questions that are not asked

Images