Analyzing the Support Level for Tips Extracted from Product Reviews

Useful tips extracted from product reviews assist customers to take a more informed purchase decision, as well as making a better, easier, and safer usage of the product. In this work we argue that extracted tips should be examined based on the amount of support and opposition they receive from all product reviews. A classifier, developed for this purpose, determines the degree to which a tip is supported or contradicted by a single review sentence. These support-levels are then aggregated over all review sentences, providing a global support score, and a global contradiction score, reflecting the support-level of all reviews to the given tip, thus improving the customer confidence in the tip validity. By analyzing a large set of tips extracted from product reviews, we propose a novel taxonomy for categorizing tips as highly-supported, highly-contradicted, controversial (supported and contradicted), and anecdotal (neither supported nor contradicted).