Research On Sentiment Analysis Of Micro-Blog User Based On DS Evidence Theory

Emotion is the attitude to that whether the external adapt to the need of individual.Our cognition,behavior,beliefs and opinions of reality are affected by emotion.The development of sentiment analysis is in step with the development of social media.In human history,we possess a wealth of data stored in digital form for the first time.These user-generated contents contain people's opinions.Mining useful knowledge from these corpora gives rise to the task of sentiment analysis.Because of its importance,research of sentiment analysis has spread from computer science to management science and social sciences.Its applications also involve many fields,such as business,politics,economy and military.In this thesis,we will research the sentiment classification in the context of micro-blog posts.First of all,this thesis puts forward a kind of emotional lexicon extension algorithm based on collaborative filtering.The algorithm makes use of seed emotional words to detect new words.By using significant emotional feature words in the existing emotional lexicon as the seed emotional words to generate weibo-words rating matrix,the method adopts singular value decomposition to build matrix.The method can reduce the computation load and retain the important information in the matrix.The TOP-N emotional words are recommended by computing the scoring matrix.Through the contrast between before and after experiment,the results show that the emotional lexicon extension can improve the coverage of emotional words in micro-blog,and significantly enhance the accuracy in the sentiment analysis.Secondly,this thesis combines a variety of sentiment analysis algorithms,and presents a sentiment classification model based on DS evidence fusion.This model treats sentiment category as the frame of discernment of evidence theory and makes use of the results of classifier to build the basic probability assignment.Finally,the results of multiple emotional classifiers are obtained by DS evidence theory fusion rules,and the final emotional categories are obtained.The results show that the method can eliminate the uncertainty caused by single classification algorithm.The accuracy of the sentiment analysis is 54.3%,which is 4.2 percent higher than single classification algorithm.