Techniques And Applications Of Sentiment Analysis On Internet User Generated Contents

The task of text sentiment analysis is to classify the sentiment orientation of natural language text, which will be helpful to understand the sentiment orientation of paragraphs and the users’sentiment orientation in a more meticulous way. Sentiment analysis can be used in several NLP applications, e.g., stock movements prediction, social network analysis and so on.With the development of information technology and social networking, people tend to use short text to express their opinions and comments, such as Twitter, microblog, mobile-phone short messages. So this work focuses on sentence level and entity and aspect level sentiment analysis on user generated contents in Internet. The sentence level sentiment analysis determines the sentiment orientation of a given sentence. And the entity and aspect level sentiment analysis is to first identify the target named entities (such as mobile phone) and/or its aspect (such as battery) and then to detect the users’ sentiment orientation on this identified entity or aspect.Regarding the sentence level sentiment analysis, this thesis adopts several supervised machine learning algorithms to build the classification model. First of all, multiple informative features, such as, semantic feature, sentiment lexicon feature and syntactic features are extracted from sentence, and then different machine learning algorithms are used to build the sentiment classification model. This is the first research work of this thesis.The second research work of this thesis is entity and aspect level sentiment analysis. The thesis focuses on the laptop and restaurant reviews. This research work contains two tasks:1) aspect term extraction and2) sentiment analysis of aspect. For the first task, we utilize Named Entity Recognition and noun phrases extraction to extract aspect terms. For the sentiment analysis of aspect, we extract multiple features from aspect context and use supervised machine learning method to learn a classifier for sentiment classification.Experimental results on SemEval2013and SemEval2014datasets show that Named Entity Recognition has better performance on aspect term extraction than noun phrase extraction. In addition, using a variety of features and supervised learning algorithms is able to effectively detect the sentiment orientation in sentence level and entity and aspect level sentiment analysis.