Research On Semantic Understanding Technology Of Question Based On Knowledge Graph

With the rapid development of the Internet,how to quickly and accurately obtain the information people need in the vast amount of data is the direction of Web development.Accurate understanding of the user's natural language questions and then retrievaling in the Web data or knowledge graph,is one of the core components of the next generation of intelligent search engine.Aiming at the question semantic understanding in search engine or question answering system,this paper designs a question semantic understanding system based on knowledge graph,which transforms the user's question into the structured graph database query and then obtain intelligent answers.The main work of this paper is as follows:This paper proposes a novel key phrase extraction method based on contextual features and XGBoost.In order to extract high-quality key phrases from questions,this paper proposes a key phrase extraction method based on contextual features and XGBoost.The method firstly constructs the phrase dictionary by means of Baidu encyclopedia with rich phrase link information,then extracts the candidate phrase in each question through the pre-constructed phrase dictionary and then extracts the rich phrase context features to train the XGBoost classifier.Last,the trained classifier is used to extract the key phrases.Experiments show that the proposed method of extracting key phrases improves the accuracy of key phrase extraction.In this paper,the task of semantic parsing of the question is treated as a task of candidate attributes ranking.This paper designs and implements the candidate attributes ranking method based on the traditional convolutional neural networks and the improved attentive max-pooling CNN.In the attentive max-pooling CNN mothod,a simplified attention mechanism is introduced in the pooling layer so that the attribute information from the knowledge graph can affect the distributed representation of questions and then improve the accuracy of the candidate attributes ranking.