An assessment of machine learning techniques for review recommendation

In this paper, we consider a classification-based approach to the recommendation of user-generated product reviews. In particular, we develop review ranking techniques that allow the most helpful reviews for a particular product to be recommended, thereby facilitating users to readily asses the quality of the product in question. We apply a
supervised machine learning approach to this task and compare the performance achieved by several classification algorithms using a large-scale study based on TripAdvisor hotel reviews. Our findings indicate that our approach is successful in recommending helpful reviews compared
to benchmark ranking schemes, and further we highlight an interesting performance asymmetry that is biased in favour of reviews expressing negative sentiment.