Rule Based System for Enhancing Recall for Feature Mining from Short Sentences in Customer Review Documents

Abstract- This paper discovers rules for enhancing the recall values of sentences containing opinions from customer review documents. It does so by mining the features and opinion from different blogs, news site, and review sites. With the advent of numerous web sites which are posting online reviews and opinion there has been exponential growth of user generated contents. Since almost all the contents are stored in unstructured or semi-structured format, mining of features and opinions from it has become a challenging task. The paper extracts features and thereby opinions sentences using semantic and linguistic analysis of text documents. The polarity of the extracted opinions is established using numeric score values obtained through Senti-WordNet. The system shows that normal rules discovered earlier are not sufficient to improve recall values as some of the opinions does not contain sentences which are linguistically correct but they express the main idea what the writer wants to convey about his opinion on a particular product. Our experiment uses a method which first identifies short sentences and then uses rules which can be applied on those sentences so that the recall values are enhanced. The paper also applies rules on sentences which are linguistically and syntactically incorrect. The efficacy of the system is established through experimentation over customer reviews on four different models of digital camera, and iPhone