Aspect selection for social recommender systems

In this paper, we extend our previous work on social recommender systems to harness knowledge from product reviews. By mining product reviews, we can exploit sentiment-rich content to ascertain user opinion expressed over product aspects. Aspect aware sentiment analysis provides a more structured approach to product comparison. However, aspects extracted using NLP-based techniques remain too large and lead to poor quality product comparison metrics. To overcome this problem, we explore the utility of feature selection heuristics based on frequency counts and Information Gain (IG) to rank and select the most useful aspects. Here an interesting contribution is the use of top ranked products from Amazon to formulate a binary classification over products to form the basis for the supervised IG metric. Experimental results on three related product families (Compact Cameras, DSLR Cameras and Point & Shoot Cameras) extracted from Amazon.com demonstrate the effectiveness of incorporating feature selection techniques for aspect selection in recommendation task.