Question Answering System Model Based on Multi Granularity Semantic Representation

Abstract

Aiming at the problems of polysemy and overflow words in traditional vector representation methods such as word2vec and glove in question answering system, this paper introduces Elmo language model, char-CNN and traditional natural language features, and makes a complete semantic representation of the original corpus text from multiple language granularity levels A neural network model for multi granularity semantic representation. Experiments on the squad dataset show that the proposed model is superior to the classical bidaf model in terms of accurate matching rate and F1 value.