Method Based On Paraphrasing Chinese Feelings

With the rapid development of the social networks, especially the rise of Web2.0,an expanding number of people begin to express their comments and views on thenetwork about events or products. As the most important source of emotionalinformation, product reviews have been widely used in academia. How to explore theabundant information on these emotions has become a hot topic in the field of thenatural language processing (Natural Language Processing, NLP).Paraphrases are sentences or phrases that convey the same meaning using differentwords. As paraphrase’s resources can be accessed generally and more linguistic featurescan be introduced at the same time, paraphrase has played a key role in many NLPapplications. This paper proposes an effective approach to recognize opinion paraphraseby exploring the characteristic of opinion paraphrase, which combined semantic andsentiment polarity. On this basis, we extend the product reviews by paraphrasegenerating technology, so as to improve the quality of sentiment analysis. Specifically,our research concerns the following three aspects:This paper presents a morpheme-based method for Chinese unknown words senseprediction. Chinese word sense tagging is the basic task of semantic analysis, which canprovide effective features for other Chinese information processing tasks. Unknownwords sense prediction has been restricting the Chinese word sense disambiguationaccuracy. This paper takes morpheme as basic unit to explore both internal and externalfeatures, and combine them for sense prediction of Chinese unknown words in theframework of maximum entropy model. Our experimental results show that theincorporation of internal and external features is of great value to Chinese unknownwords sense prediction.Chinese opinion paraphrase identification by combining semantic analysis andsentiment polarity classification. This paper tries to investigate the integration ofsemantic feature and sentiment polarity for opinion paraphrase identifying; we alsoanalyze and compare the different paraphrase recognition methods. Experimental results demonstrate the effectiveness of the integration of semantic and sentiment polarity.Sentiment orientation analysis based on opinion paraphrase generation. This paperputs forward a paraphrase-based method to expand opinion corpus, so as to solve theproblem of data sparse that statistical methods are facing. We conduct differentsentiment classification experiments and analysis the results under different conditions,experimental results show that the introduction of paraphrase generation can improveclassification accuracy significantly.