Document Sentiment Analysis Based On Sentence Sentiment Weight Synthesis Algorithm

Along with the vigorous development of the internet, more and morepeople express their views on some hot events through some networkcommunities, such as micro-blog, forum and circle of friends. And consumersalso show their views on some goods through the network frequently. Thesecomment texts which include subjective emotion of users contain the users’real idea directly. So it has great social and commercial values to analyze thoseweb comment texts effectively. Processing a large scale web text data artificiallyneed to consume lots of manpower and resources. While using highperformance computer technique to analyze the web text can increaseefficiency greatly. And using computer to analyze the emotion of text canextract the useful information effectively. Traditional text sentiment analysis ismainly to analyze words, phrases and sentences. Research on text of lengthy isrelatively poor. And some special complex sentences in Chinese are lack ofenough analysis. Therefore, this study focuses on text of lengthy as the mainresearch object and analyzes the structure features of complex sentences.Finally, this paper proposes a new text sentiment analysis method based onsentence sentiment weight synthesis algorithm to analyze the sentimentorientation of lengthy Chinese text effectively.This study mainly accomplished the following works. Firstly, goodrecognition of words was the key to future research. So in order to identifyvarious related words better, emotion words dictionary, conjunctions table,negative words table, summary words table, degree adverbs table and so onwere constructed on the basis of existing resources. And they were integratedinto word segmentation to improve segmentation accuracy. Secondly, this study fully analyzed the traditional na ve bayes text classification algorithm andimproved it. Then used the improved algorithm to analyze the sentiment ofsimple sentences, and compared the new algorithm with the traditional na vebayes algorithm finally. Thirdly, this study fully analyzed the special sentencestructure of complex sentences in Chinese. Because of the situation thatvarious semantics co-occurred in complex sentences, using traditional textclassification algorithm could not analyze the sentiment orientation of complexsentences effectively. So this study proposed a sentiment analysis rule forcomplex sentences according to the combination rules of conjunction, negativeword and emotion word. Finally, this study proposed a sentiment analysismethod for lengthy Chinese text. When analyzing lengthy text, the whole textwas segmented into a set of sentences. Then this study used differentsentiment analysis method to analyze simple and complex sentences. Aftergetting the sentiment polarity, each sentence was given a sentiment weightaccording to the special factors which affect the emotion of sentence, such asdegree adverbs, sentence patterns, and sentence position and so on. Aftergetting the sentiment weight of each sentence, the final sentiment polarity ofthe whole text could be got according to the sentence sentiment weightsynthesis algorithm.In the final experiment, the improved na ve bayes algorithm hadbetter accuracy than traditional na ve bayes algorithm. And using complexsentence sentiment analysis rule to analyze the emotion of complex sentenceshad better analysis effect than traditional text classification algorithm. Finally,when using sentence sentiment weight synthesis algorithm to analyze threekinds of document, the average accuracy rate was80.6%,81.4%, and74.6%andthe average recall rate was80.1%,82%, and77.2%. However, when usingtraditional text classification algorithm to analyze the emotion of samedocument, the average accuracy rate was73.4%,76.2%, and70.5%and theaverage recall rate was76.1%,78.3%, and72.3%. Experiment result proved thatusing sentence sentiment weight synthesis algorithm could analyze thesentiment orientation of document effectively and had higher efficiency than traditional text classification algorithm.