Automatic acquisition of feature-opinion pairs in customer reviews

Customer reviews mining could urge rational consumption and production. In this article, we extracted the product features and opinion words in a unified process with semi-supervised learning algorithm (Bootstrapping), obtained the product features in which customers are interested. Then we dealt with the comments with multi-features but single-opinion with the obtained feature words and opinion words. For semi-supervised learning algorithm would cause a sharp decrease in precision with the iteration times, we maximized the harmonic-mean to adjust the sequences of features and opinions with big standard deviation, and deleted the sequences with low-frequency, thus not only increased the precision but also ensured the recall. The experiment results show that it was very effective. 1548-7741/ Copyright © 2009 Binary Information Press.