A Product Aspects Identification Method by Using Translation-Based Language Model

Aspect identification is a fundamental basic process for web opinion analysis of product reviews, which shows great potential in many real applications of e-commerce. The main stream of the solutions of aspects identification is to explore the modification relationship between aspect and opinion words. In this paper, we investigate the complexity of the review sentence to better capture the modification relations. Then, we propose a language model based aspect identification method integrated with a translation model to exploit these modification relations for performance improvement. The experimental results on 11 popular products in four domains from various web sites show our approach is more effective compared with some strong baselines and the state-of-the-art methods.