The Study Of Vertical Search Engine Based On The Product Review Of Classified Sentiments

With the development of Internet technology, e-commerce, andcyber-network products such as BBS, blog, and MicroBlog, the BTOCexchanges have become more and more common.Nowadays more and more customers put their product reviews on theInternet after their use. Likewise, the number of reviews is increasing,among which most are colloquial and nonstructural.It costs much for business people to select self-concernedinformation manually from the massive product reviews to makedecisions about the supply-demand relations. That’s the same case forpotential customers to buy products for it is one-sided and lagging.Thus search engine has played a significant role in the Internet. Butthe famous search engines like Baidu and Google cannot fully meet thedemands of researches in specific product reviews, which are generalsearch engine for different products in different domains.In such a circumstance, the study and development of a verticalsearch engine based on the product review of classified sentiments isdefinitely necessary.On the basis of researches and studies in and abroad, this thesis hasmade a further study on the recognition of comment target in commentsof Chinese products, evaluation phrase, collocation of comment targetand evaluation phrase, and on the analysis of sentimental orientation ofevaluation phrase. The main work is as follows:（1）Part of speech sequence templates are used to extract candidatecomment targets during the progress of recognizing the commenttargets.The comment targets’ integrity and stability and algorithm areproposed to filter the useless words in the comment targets.The commenttargets and evaluation phrases’ co-occurrence rules and the frequency ofcomment targets in the whole review are used to sort the confidence levelof comment targets.And finally comment targets are extracted accordingto the confidence level.（2）The conjunction dictionary,sentiment dictionary,degree wordsdictionary and negative words dictionary are improved to recognize comment phrases and analysis the sentiment orientation of commentphrases.There are eight features which are about the relationship betweencomment targets and comment phrases.These features and the algorithmof SVM are used to recognize the collocation of comment targets andcomment phrases.Ultimately,the sentiment orientation of whole review ofcomment text is determined by these methods above.(3) The popular framework SSH, the database software mysql and theopen source of lucene are uesd to develop a vertical search engine whichis based on sentiment orientation of procuct reviews.And users canconveniently search the information they are interested in.The above studies can help business people and potential customers tocarry out their market research or make purchase decisions in aconvenient way.This study method about classified sentiments is both of commercialvalue and of academic value.