Research On The Technology Of Sentiment Orientation Analysis In Chinese Web Text

The internet is becoming an important media for the public to get information benefitfrom the development of the network technology. However, the people not only just getinformation from the Internet nowadays, but also want to express their viewpoint, attitude andopinion about any object and events. Especially in the past several years, a lot of content withSentiment orientation appeared on the Internet, with the development of blog, BBS, microblog, electronic commerce, social network. Informed the word of mouth about the publiccommodity towards a commodity can help the businessman make a beneficial decision.Mining the attitude towards social events, master the public opinion will be benefit to thegovernment maintain the social stability. So the research of sentiment orientation analysisabout text has great mining.The sentiment orientation analysis about text has a wide field. In this paper, weintroduce the key technology that sentiment orientation analysis involves in great detail,include the standards of classify, Chinese Segmentation, research of orientation words, thetechnology of text classify in orientation, the technology of disambiguation about words, andso on. We do some research in great detail about these key technologies used at home andabroad.Nowadays, about the research on sentiment orientation in whole article level, the recalland precision is not good enough. So in this paper, we present a new approach in constitute asentiment orientation dictionary based on HowNet and Synonyms Thesaurus. And then wepresent a new method to analyze sentiment orientation with multiple features based on thesentiment orientation dictionary we constituted. In this method, we take the transitionalcomplex sentences, negative words, degree adverb words into consideration. We didexperiments use the public data source to test the method presented in this paper, and theresult show good recall and precision.In different context, the same word can show different sentiment orientation, we callthese words dynamic sentiment words. One difficulty of text sentiment orientation based onsemantic is the process of dynamic sentiment words. In this paper, we present a new approachto deal with the dynamic sentiment words based on Bootstrapping algorithm. In this method, we use a small number of corpus labeled by hand, extend the scale of the seed corpus viarepeatedly iteration. And later, we disambiguated about the dynamic sentiment words usemachine learning. In this paper, we do experiments about this method on six dynamicsentiment words, and achieved the intended purpose.