Research On Sentiment Orientation Analysis Of Product Comment In Evaluation Objects

The users always explain their subjective attitudes on the components, performance and function of products in their review text after purchasing products. The review text greatly affects the users’ purchase decision. Thus, the research on users’sentiment orientation on the products in the review text is very significant. We analysis a large number of review text on products, and discovery that the users not only concerned about the overall quality of a product, but also pay more attention to the product features. In this paper, we analysis the users’ sentiment orientation based on product features from the review text.This paper focuses on the critical work on the text sentiment analysis. From the laptop reviews corpus, we extract specific evaluation objects and the evaluation words, and analysis the relationship of these evaluation objects and the evaluation words so that we can obtain their tuples. Then, for every evaluation object, we analysis the reviewers’ sentiment orientation in the feature level through the review text. Our works are mainly consisted of the following contents.First, research on the extraction technology of the evaluation objects, evaluation words and their collocation relationship in the products reviews. Combining knowledge-based methods and machine learning methods, we build the knowledge library based on the field of laptop, and extract the common evaluation objects. Then build a model using the Conditional Random Fields Model (CRFs). To extract those new evaluation objects, the word feature and the word class feature are blended into the CRFs. The evaluation objects which have been extracted are important for our rest work. We add the modification word feature and location feature into the previous model, and extract evaluation words and their collocation relationship. The precision and recall of our methods have been verified in our experiments.Second, researching the feature ontology construction methods on the laptop field based on evaluation unit. Taking into account that the evaluation objects are not independent, they are interconnected each other. We build the feature ontology of the laptop. The tuple<evaluation object, evaluation words> are added into the feature ontology as the property concepts, which can be used to assisted sentiment orientation analysis.Third, we proposed a sentiment orientation analysis method based on emotional vocabulary. Improving the existing emotion dictionary, and the evaluation dictionary and emotion dictionary has been constructed respectively. Considering the characteristic of sentiment in the product reviews, we divided the users’sentiment into four categories, and analysis sentiment orientation using a fine-grained method by calculating the similarity of the words. Meanwhile, we consider those descriptive words which are always targeted to the evaluation object. Therefore, we build the descriptive words set to assist the sentiment analysis, which can improve the precision sentiment analysis. In the experiment, the effectiveness of the proposed method is verified.