Research on the Automatic Evaluation of Merchandise Comments on Blogs

Opinionated content in Blog comments usually has a positive or a negative or a neutral connotation. This paper researches on the automatic evaluation of Blog comments on merchandise. It builds a meta-search engine for retrieving Blog pages with merchandise comments. The retrieved pages are parsed; the comment texts are drawn out and serve as resources for corpus. By means of feature extraction and polarity analysis, these comments text are represented with SVM. In polarity analyzing, a dictionary of merchandise attributes and a lexicon of positive and negative words are constructed to improve the accuracy. The average score of specific product is calculated based on the polarity score of each merchandise attribute and the percentage of people who gives positive comment. We build a prototype system AESBC and conduct comparison study between the result of our experimental system and that of field experts. The experimental result shows the effectiveness of our method.