The Domain-Dependent Noun Polarity Analysis Based On Feature Extraction Technique

In social networks, for the huge number of products and services, more and more online unstructured reviews are generated every day. Online users not only express an overall opinion on the product and service, but also express their specific feelings for product features of interest. In the feature level, automatically identifying users’opinion can provide producers and consumers much support on decision-making. Although there are many feature level opinion mining methods, only a few study focus on objective and emotionless noun feature which express positive or negative opinions in particular fields of knowledge.In this paper, we analyzed the feature level domain-dependent noun polarity. For noun features, we used the feature extraction algorithm based on linguistic framework. For each sentence in the text, we were implementing word processing and the corresponding part of speech tagging. Next, it is very important to determine the dependence relationship on each word with other words, as well as the scope of each word. In this way, we can accurately extract noun features from the sentence. Meanwhile, in order to calculate the sentiment score subsequently, we also need to extract the adverb-adjective-noun phrase structure.In opinion mining, mining domain-dependent opinion word is a very important issue. However, the existing work analyzed the adjectives and verbs, only a little work focused on nouns and noun phrases. In our study, we used an opinion mining method based on linguistic analysis to identify and extract objective noun feature expressing opinions in specific domains. First, we proposed a method for automatic noun feature extraction. This extraction method used linguistic framework to extract adverb-adjective-noun phrase structure. Subsequently, we propose a general additive model to calculate the sentiment score for noun features. Then, by using a statistical method and deleteing noun features which don’t express opinions in a particular area, we can determine the objective noun features which imply opinions in specific domains. Finally, we conducted several experiments to present the advantages of my proposed method. Based on real life datasets, the proposed method shew advantages on accuracy and robustness.In addition, our results had the expected recall and precision.