Modeling User-generated Content with Item Response Theory
In this project, we study a new text mining problem that aims at decoupling user's personalized topical representation and item's intrinsic properties from user-generated contents. Without properly decoupling the interdependence between user and item, one can hardly obtain accurate user understanding and item profiling. Motivated by the Item Response Theory (IRT) from psychometric studies, we propose to model user-generated content as a user's response to an item, and assume the response is jointly determined by the individuality of the user and the property of the item. The process is modeled with a generative topic model, in which we decouple the items' intrinsic properties and users' manifestations of them in a low-dimensional topic space. The learned user and item representations after decoupling will enable improved performance in many practical tasks, e.g. personalized recommendation, friend suggestion.
Committee: Alf Weaver (Chair), Hongning Wang (Advisor), Farzad Farnoud, Yangfeng Ji