A Research Of Sentiment Analysis In User Comments Mining

The more Internet data grows, the more potential applications it shows.Data Mining, such as sentiment analysis, is a technique to find someparticular patterns in big data, it therefore can act as a guidance in people’sliving and production. Sentiment is a high-order semantic confined to naturallanguage, whose mining needs lots of labor. But sentiment analysis introduceautomation and intelligence in sentiment mining and human-computerinteraction.There are there level-specify tasks in Sentiment Analysis: sentimentextraction, sentiment classification, and sentiment retrieval. Traditionalresearch has the following limitations: firstly, sentiment extraction eitherdoesn’t take into account the grammatical structure of text, or resort to somecomplex model, which is hard to train; Secondly, sentiment classificationinvolves so much heuristic features of text, that its adaptability is poor;Thirdly, sentiment retrieval on text doesn’t take into account the staticquality of text, which is crucial in the traditional information retrieval.This paper is dedicated to user reviews, which are characteristic by singlepublisher and single subject. Taking advantage of those features of userreviews, this paper manage to solve those problem mentioned above in thefollowing way. Firstly, this paper introduce new model in sentimentextraction, which both involve the grammatical relations and is easy to train.Secondly, this paper reduce every other classification features into sentimentpair, which is the only text features used in sentiment classification; Thirdly,to take into account the static quality of text on sentiment retrieval, this paperinvolve the user reputation mechanism. In sum, this paper introduce some new model and algorithm on sentiment analysis, and justify them bybenchmark test.