查看文章

Explanations are an important part of modern recommender systems. They help users to make better decisions, improve the conversion rate of browsers into buyers, and lead to greater user satisfaction in the long-run. In this paper, we extend recent work on generating explanations by mining user reviews. We show how this leads to a novel explanation format that can be tailored for the needs of the individual user. Moreover, we demonstrate how the explanations themselves can be used to rank recommendations so that items which can be associated with a more compelling explanation are ranked ahead of items that have a less compelling explanation. We evaluate our approach using a large-scale, real-world TripAdvisor dataset.