Sentiment Polarity Level Analysis Based On Textual E-commence Comments

Sentiment analysis is a study field addressed on analyzing people’s opinions,sentiment, evaluations, appraisals, attitudes, and emotions towards entities such asproducts, services, organizations, individuals, issues, events, topics, and their attributes.It involves a number of emotional and challenging research tasks. According to thedifferent tasks, Sentiment analysis involves sentiment classification, sentimentextraction, opinion search and opinion summary. Research process includespretreatment, sentiment information extraction, classifier selection, the resultssummarization and show.With the development of Web2.0, people can submit their views via the Internetanytime and anywhere, such as blog, forums, portals, ecommerce platform. These textscan directly or indirectly reflect human’s behavior and thoughts, so it is useful toanalysis these texts. Concerned about the sentiment analysis of e-commerce commenttexts, both for consumers and businesses have important significance. Although thecurrent sentiment analysis has made great progress, the level of sentiment polar researchalso insufficient. There is inherent fuzzy characteristic in texts, which can calculationpolarity level to take advantage of the membership function of the fuzzy theory.The main task of sentiment analysis is to determine the level of polar and polar oftext, that the strength of belong to the positive or negative. According to thecharacteristics of e-commerce comment text, based on the existing traditional sentimentanalysis, to carry out analysis of the level of polarity in comment sentence. First, thesentiment of adjectives in WordNet dictionary is marked by STEP algorithm. And thencalculate the polar level of word by NOS membership function. Secondly, the domainsentiment dictionary was achieved through transformation NOS sentiment dictionary bythe use of integer linear programming (ILP). So that e-commerce dictionary canmaximize adaptation text analysis. Then, to improve the traditional weighted statisticalalgorithm, we use a new method to judge the polar of sentences in advance by electoralvote, and then calculate an average value of trandational method and new method.Finally, use the dictionary to extract the domain sentiment of text, including sentimentdictionary, negative words, and degree adverbs. And create an e-commerce sentimentclassification system.Experimental corpuses have12,000comment sentences, which derived from e-commerence websites, include household appliances, hotel, book, respectively. Thepolar through manual annotation as the gold standard for test result. Experimentalresults show that: the average accuracy rate of the domain sentiment dictionary thanNOS dictionary increased2.7%～6.1%; the improved algorithm than the weightedstatistical algorithm improved5.7%～9.1%, indicating that the initial attempt tosentiment analysis conducted have better results.