| How to quickly and accurately obtain the product reviews on the Internet andanalyze its semantic orientation has great significance both to Business’ graspingconsumer preferences and potential consumers making purchasing decisions. However,facing the abundant reviews on the Internet, it’s a time-consuming process thatscreening and generalizing a great capacity of information manually. Therefore,research about sentiment analysis of text comes into being, and it has been a researchhotspot on Natural Language Processing.In this paper, the researches focuses on the key technologies about sentimentanalysis, such as, recognition about evaluation object and sentiment words,distinguishing orientation of sentiment words. According to the existing research, thispaper made some attempts to exploring the concrete modes and effects provided bydomain ontology, and achieved this tasks combined with statistics and semantic analysis,The thesis completes the following tasks:1. Against at most of the rule-based evaluation object extraction can only find thefrequent evaluation objects at present, and the shortcoming of extracting the infrequentevaluation objects with low accuracy, this paper takes the car domain as an example, inorder to overcome that statistical method lacks of semantic information between wordsin domain concepts extraction, a combined domain concept automatic extractionmethod is presented, and then the Car Ontology is built by using Protégé. Pair ofevaluation object and sentiment word extraction algorithm based on domain ontologyand SBV is put forward, which is called I-SBV algorithm, so that the accuracy ofextracting evaluation objects is improved and the identification of product attributerelations can be achieved. 2. Considering the existing Sentiment Lexicon ignores that the evaluation objectsimpact on sentiment polarity in the build process, at the same time, according to theacronyms and abbreviations often appears in the web language, an Sentiment Lexicon incar field is built, which including three parts, Static Sentiment Lexicon which integratesthe network buzzwords, Dynamic Sentiment Lexicon and Modifier Lexicon. thepolarity value of the no-login sentiment word is calculate by using extension SO-PMI,meanwhile, the calculation formula of dynamic sentimental word polarity is given,which provides a good foundation of the realization of the sentiment analysis based onthe sentiment lexicon.3. Based on the Domain Sentiment Lexicon, and considering that modifier wordshave influences on sentiment word polarity, a calculation method of sentimental wordcontext polarity value is given. After that, taken the emotional phrases and evaluationobjects as the basic unit for calculating the polarity of sentence, the sentence polarity isequal to weighted sum of emotional words polarity values, which takes advantage of therelationships between ontology concept and their attributes. Finally, the task of sentencetendency discrimination is completed.Based on the above research, the Sentiment Analysis System about car reviews isdesigned and implemented. |
PDF Full Text Request