Hidden factors and hidden topics: understanding rating dimensions with review text

摘要: In order to recommend products users we must ultimately predict how a user will respond new product. To do so uncover the implicit tastes of each as well properties For example, in whether enjoy Harry Potter, it helps identify that book is about wizards, user's level interest wizardry. User feedback required discover these latent product and dimensions. Such often comes form numeric rating accompanied by review text. However, traditional methods discard text, which makes dimensions difficult interpret, since they ignore very text justifies rating. this paper, aim combine (such those latent-factor recommender systems) with topics learned topic models like LDA). Our approach has several advantages. Firstly, obtain highly interpretable textual labels for dimensions, us `justify' ratings Secondly, our more accurately predicts harnessing information present text; especially true users, who may have too few model their factors, yet still provide substantial from even single review. Thirdly, discovered can be used facilitate other tasks such automated genre discovery, useful representative reviews.