Structured Methods For Opinion Mining

Sentiment analysis and opinion mining have received much attention in re-cent years. A number of automatic methods have been proposed to identify and extract opinions, emotions, and sentiments from text. It will facilitate both opin-ion related application and other natural language processing(NLP) tasks. Struc-tured learning, which utilize structure information to improve machine learning approaches, has successed in many NLP field and is considered to be one of the most effective methods.This work is focusd on structured learning methods in opinion mining. First we present a novel approach for mining opinions from product reviews, where it converts opinion mining task to identify product features, expressions of opin-ions and relations between them. Previous works on this topic are either simply relate adjacent opinion expression and product feature, or use hand-written pat-terns to extract relation directly. By taking advantage of the observation that a lot of product features are phrases, a concept of phrase dependency parsing is introduced, which extends traditional dependency parsing to phrase level. This concept is then implemented for extracting relations between product features and expressions of opinions by a newly designed tree kerenl. Experimental eval-uations show that the mining task can benefit from phrase dependency parsing and the new tree kernel function.Second, Based on analysis of on-line review corpus we observe that most sentences have complicated opinion structures. Existing methods, such as frame-based and feature-based ones, igore a lot of useful information. In this work, a novel graph-based representation for sentence level sentiment is proposed. An structured learning method with integer linear programming-based inference al-gorithm is then introduced to produce the graph representations of input sen-tences. Experimental evaluations on a manually labeled Chinese corpus demon-strate the effectiveness of the proposed approach.