Research On The Method Of Solving Multiple Choice Questions Of Reading Comprehension Based On Keyword Semantic Expansion

In recent years,with the rapid development of artificial intelligence,problems on solving intelligence have attracted wide attention,among which machine reading comprehension has become an important research branch in this field.Aiming at the Chinese reading comprehension task of college entrance examination,this thesis studies the automatic solution of selected topics in science and technology articles.This type of contents generally contains a background material,a question stem and four candidate options.The machine is required to select the correct option from the four candidate options according to the question stem and based on the understanding of relevant content in the background material.In this thesis,we propose a problem solving model based on keyword semantic expansion,which enriches material information through keyword semantic expansion based on external knowledge base,so as to obtain better problem solving effect.The main research contents of this thesis include:(1)Keyword's extraction based on multi-class feature fusion.According to the questions and options given by each question,the keyword extraction algorithm based on multi-class feature fusion is adopted to extract keywords,which is based on three different algorithms,including TF-IDF,Text Rank and LDAA,and comprehensively extracts keywords for text information of different dimensions.(2)Semantic expansion based on external knowledge base.The knowledge base is built according to the entry information of baidu encyclopedia.In view of the keywords extracted by the multi-class feature fusion algorithm,the semantic expansion of the keywords in the answer candidate sentences,question stem and options is carried out based on the established knowledge base,so as to enrich the text information represented by them.(3)Multiple choice questions based on keyword semantic expansion.The Bert model can capture context information better and represent richer text information.Therefore,this thesis builds a multiple-choice answer model based on the background material,question stem and options after semantic expansion,and selects answers by calculating the semantic relevance value.Meanwhile,relevant experiments are conducted on the data set of established college entrance examination multiple choice questions to verify the validity of the proposed method.