The Research And Analysis Of Automatic Summary Based On Users' Reviews

With the rapid development of e-commerce,network information is increasing.As more and more people shop online,how to improve the user experience,strengthen information exchange between users has become an important issue.Leaving comments after the user's shopping is an important way between the users and business merchant to communicate,this thesis presents the way to study a summary of consumer reviews.User comments on mining and traditional text mining is different,because the user comments generally much shorter than the text,information point finer focus.It involves a lot of natural language processing,machine learning and data mining technology.With the development of machine learning,especially the rise of the deep learning,a lot of problems have been further in-depth research.Combined with basic natural language processing,association mining algorithm,hierarchical clustering model,neural network and decision tree algorithms,the thesis do some new research on it.Aiming Chinese,improved Apriori algorithm for extracting features of English reviews,and achieved good results,proved the feasibility of the approach.Using the word activation force model to cluster the Comments characteristics,it has stronger adaptability than traditional clustering model.Based on word2vec,using the self-recursive auto-code neural networks get better results than the traditional naive Bayes classifier on sentiment analysis of Comments sentence,improved the value of F1 about 8%.Finally,using a decision tree established feature of the hierarchical model get a better organization of the summary display.