ILDA: interdependent LDA model for learning latent aspects and their ratings from online product reviews

摘要: Today, more and product reviews become available on the Internet, e.g., review forums, discussion groups, Blogs. However, it is almost impossible for a customer to read all of different possibly even contradictory opinions make an informed decision. Therefore, mining online (opinion mining) has emerged as interesting new research direction. Extracting aspects corresponding ratings important challenge in opinion mining. An aspect attribute or component product, e.g. 'screen' digital camera. It common that reviewers use words describe (e.g. 'LCD', 'display', 'screen'). A rating intended interpretation user satisfaction terms numerical values. Reviewers usually express by set sentiments, 'blurry screen'. In this paper we present three probabilistic graphical models which aim extract products from reviews. The first two extend standard PLSI LDA generate rated summary As our main contribution, introduce Interdependent Latent Dirichlet Allocation (ILDA) model. This model natural task since underlying assumptions (interdependency between ratings) are appropriate problem domain. We conduct experiments real life dataset, Epinions.com, demonstrating improved effectiveness ILDA likelihood held-out test set, accuracy ratings.