Complex question answering method and device based on knowledge graph and storage medium

Abstract

The invention discloses a complex question answering method and device based on a knowledge graph, and a storage medium. The method comprises the following steps of performing entity recognition and entity linking on an input question, obtaining a candidate key path and knowledge graph information according to the subject entity of the question, constructing a candidate critical path matching model, identifying a critical path of the question, and identifying candidate constraint conditions corresponding to the selected critical path, constructing a constraint condition screening model, and selecting a proper constraint condition, and generating a query graph of the question sentences in combination with the key paths and the constraint conditions, and retrieving answers in the knowledge graph. According to the method, all aspects of information of the knowledge graph can be fully utilized, key semantic information of complex questions can be extracted, matching of the semantic information of the questions and the knowledge graph is achieved, and the accuracy of the answers is improved by automatically screening constraint conditions, narrowing the answer range and positioning question points of the questions.

Description

technical field [0001] The present invention relates to the technical field of natural language understanding, in particular to a complex question answering method, device and storage medium based on knowledge graphs. Background technique [0002] In recent years, intelligent question-answering robots have gradually become one of the important research directions of many large companies, such as Microsoft Xiaoice, Ali Xiaomi, etc. As an important part of knowledge graph-based question-answering, it has attracted extensive attention from academia and industry. And in many directions such as the financial field, medical field, education field, etc., it is applied in the form of customer service and search engines. In actual application scenarios, users will face complex scenarios. These complex problems are also a technical difficulty that the current question answering system needs to solve. [0003] In related technologies, the mainstream methods are divided into two catego...

AI Technical Summary

Problems solved by technology

CN111611806A proposes a semantic analysis method for knowledge graph question answering. In this method, different parts of the question sentence are converted into different data nodes through a pre-built dictionary, and then converted into a query graph according to artificially defined rules. The method directly matches through the dictionary, does not make full use of the specific information of the knowledge graph, and is not flexible

CN110111766A proposes a knowledge map common sense question answering method based on staged query. In this method, the entity, constraint conditions and question structure of the question are identified through models of different stages. Although the unity of the overall structure is satisfied, But it does not match the semantic information of the question sentence with the parts of the query graph

[0005] 1. The existing method directly matches the relationship between the question sentence and the knowledge map, without using other information of the knowledge map, resulting in inaccurate matching

[0006] 2. When the existing method sorts candidate query graphs, it is necessary to manually define many features, convert the query graph into a feature vector and then sort through the model, and the ranking model is only for the characteristics of the query graph, and cannot match the query graph with the specific problem. one-to-one correspondence

[0007] 3. In the process of constraint identification, some methods convert different constraint items into the same feature vector, and the model cannot recognize these different constraint items

Images