Study On Opinion Mining Based On Semantic Analysis

With the continuous development and popularity of network technology, more and more Internet users post their opinions and perspectives about news events, products, services through diversities of web applications, such as the review function, blog, micro-blog and so on. Thus, the Internet has aggregated a mass of users’opinions. Since the end of last century, scholars in the fields of computer science, computer linguistic, cognitive psychology, linguistic and behavioral science have begun the study of automatic approaches to search, analyze and mine users’opinions in the Internet. This research is known as Opinion Mining. Opinion Mining is currently a major research in the intelligent processing of Internet information, which is of great practical value. It can be widely applied to information retrieval, information filtering, E-Commerce, network public opinion monitoring, spam mail handling, network user behavior tracking and some other areas.By referring to the research framework of text mining, this doctoral dissertation attempts to mine opinions from network users’reviews. Applying topic model, syntax parse, compositionality principle and appraisal theory, it is focused on following four key points:topic extraction from reviews, topic co-reference resolution, sentiment analysis and ranking the review data.25 figures and 17 tables are attached. This doctoral dissertation is divided into seven sections with about 130,000 words. The main contents are as follows:Chapter 0 is organized by the four basic subtasks of Opinion Mining. It respectively summarized the existed research ideas, theoretical methods, technical routes and trends of opinion holder identification, topic extraction, claim selection, sentiment analysis, as well as opinion mining experimental system. On this basis, the research tasks and research ideas, which lay the foundation of this paper, are come up with.Chapter 1 illustrates the theoretical basis of Opinion Mining. After analyzing the concept of Opinion Mining, the research framework of Opinion Mining is pointed out. Affect computing theory is its theoretical source; text analysis and text classification is its technology base; text mining is its methodology base. And also, respective conclusions are made about the contents of the foundational theories, such as affective computing, emotion expressing, text affect computing method, lexical analysis, syntactic analysis, text emotion recognition methods, text classification steps and text classifiers.Chapter 2 concentrates on the automatic topic extraction from reviews. This chapter first reviews the concepts and tasks of topic extraction from reviews. After summing up the basic ideas and implementing the algorithms of probabilistic models, the topic model hPAM is then proposed to extract topics from reviews. What’s more, the semantic information amongst discrete topics is fetched to generate hierarchical topics of reviews.Chapter 3 is about the research of text segmentation for review data. In this chapter, at first, reference phenomena in reviews are analyzed, with 11 reference features found. Then, SVM approach is taken to resolve topic co-reference. Lastly, Labled_LDA algorithm is functioned to segment review text in this paper.Chapter 4 makes a detail research into sentiment analysis of reviews. This chapter builds a static polarity dictionary. Thus, it determines priori polarity, polar intensity and polarity weights for words. Besides, it also concludes 6 basic principles for polarity judgment of word matches. Then,21 polarity judgment principles of verb phrases are established in accordance with the above principles. Based on these principles, this chapter therefore accomplishes polarity classification for sentences and sentiment intensity analysis by putting syntactic analysis tools as well as the principle of composition into effect.Chapter 5 shows basically the research into the issue of ranking review data. On the basis of appraisal theory framework of systematic functional linguistics, the reviews are mainly fallen into three categories:affection, judgement and appreciation. Hence, review data are ranked by appraisal expression, sentiment intensity and frequency of emotional expression.Finally, a summary of this research is listed, pointing out the limitations, and making a research prospect.