| Customers and producers have already reached a consensus on the importance of gettinginformation from product reviews. Product reviews mining has also become a hot researchtopic. A number of comparative reviews are included in product reviews. As a commonexpression, comparative sentences can help critics describe the difference of quality andservice between two products. Therefore, the study on such reviews is important.This paper studies the comparative relations among electronic products reviews, and themain work includes the following aspects:When identifying the comparative reviews, a comparative feature lexicon and asuperlative feature lexicon is firstly constructed. Then the study classifies the words in thecomparative feature lexicon according to their polarities, laying a foundation for thefollow-up.This study considers the identifying as a classification question, the “comparative” and“non-comparative”. Giving the comparative reviews usually following some fixed patterns,this study applies the class sequential rules to the first identification. For the candidatecomparative reviews and the reviews without any classes after the first identification, a newidentifying algorithm based on the product naming rules is proposed. Judging by the differentnumbers of products included in reviews, different methods are adopted for the secondidentification.During the extracting, the study uses5-items to express the mining results. Firstly, use amethod based on calculating the similarity between the brands of entities and the relatedinformation got from web pages to identify the comparative subjects and objects. Then, thispaper constructs a few extracting patterns as well as a comparative result calculating formula,which can be applied to the comparative reviews. Finally, exact the5-items successively.The experimental results show that, the method explored in this study to identify andmine the comparative reviews can improve the precision and recall to a certain extent. |
PDF Full Text Request