Research On Domain Adaptive Chinese Sentiment Lexicon Construction

With the rapid development of Web2.0, user’s behavior on the Internet ischanging dramatically, more and more people express personal view on theInternet to participate in the creation of Internet resources. Explosive surge in theInternet the amount of information, relying solely on manual methods to collectand collate the information is far from being able to meet the needs. Therefore,from the flood of Internet information automatically excavation and finishing thesubjectivity of information is particularly important task of sentiment analysis. Inrecent years, there have also been more and more the attention of scholars andresearch institutions.Sentiment analysis is an important topic of natural language processing, andseeks to analyze the structure of subjectivity text, processing and induction, andultimately the formation of the convenient machines to understand and user-structured data. Sentiment analysis is a multi-disciplinary field of study, coveringmulti-language analysis techniques, in accordance with the different textgranularity can be divided into chapter-level sentiment analysis, sentence leveland word level. Emotional words is an important part of emotional analysis, a lotof upper sentiment analysis tasks such as the document level, sentence level needthe support of sentiment dictionary. Thus, how to build high-quality sentimentdictionary is particularly important for sentiment analysis. Word has differentsentiment expression in different domains, or even possible to express theopposite polarity. It is difficult to meet all the requirements to build a complexsentiment dictionary.In this paper, we proposes a semi-supervised methods for sentimentdictionary building framework. The algorithm process consists of three steps,namely, access to seeds of emotional words to construct a semantic graph andcalculate the sentiment score. Sentiment word seeds are automatically obtainedfrom large-scale user reviews; the construction of the semantic graph depends onthe external semantic resources, such as a synonym for the CiLin. We tried theTopic-Sensitive PageRank and label propagation algorithm to calculate theemotional score.What’s more, we combined supervised learning method andRandomized Minicuts to identify the word polarity. Experimental results show the effectiveness of our method.What’s more, we proposed a domain-adaptive method for constructingsentiment dictionary based on statistical analysis. The main process containstarget extraction, the polarity of the sentiment word in the field of extraction andpolarity identification. The target extraction module use a statistical analysis ofthe combination of the Internet expansion; sentiment word extraction moduleutilize the relationship between the sentiment word with emotional path,including syntactic path and plane path to build the emotional path of thetemplate.Finally, polarity is recognized with the cosistency of intra-and intersentiment expression.Finally, this paper design and the a domain-adaptive sentiment dictionaryextraction platform SWMine, which include the areas of target extraction, thefield of sentiment word extraction and polarity identification. We also design thedata representation and the visualization platform, which can give sellers andbuyers intuitive helpness.