Representation Learning for cold-start recommendation - Poster ICLR 2015

Full Paper :

Poster : poster


A standard approach to Collaborative Filtering (CF), i.e. prediction of user ratings on items, relies on Matrix Factorization techniques. Representations for both users and items are computed from the observed ratings and used for prediction. Unfortunatly, these transductive approaches cannot handle the case of new users arriving in the system, with no known rating, a problem known as user cold-start. A common approach in this context is to ask these incoming users for a few initialization ratings. This paper presents a model to tackle this twofold problem of (i) finding good questions to ask, (ii) building efficient representations from this small amount of information. The model can also be used in a more standard (\textit{warm}) context. Our approach is evaluated on the classical CF problem and on the cold-start problem on four different datasets showing its ability to improve baseline performance in both cases.


We have proposed a new representation-based model for collaborative filtering. This inductive model (IAM) directly computes the representation of a user by cumulative translations in the latent space, each translation depending on a rating value on a particular item. We have also proposed a generic formulation of the user cold-start problem as a representation learning problem and shown that the IAM method can be instantiated in this framework allowing one to learn both which items to use in order to build a preliminary interview for incoming users, but also how to use these ratings for recommendation. The results obtained over four datasets show the ability of our approach to outperform baseline methods.
Different research directions are opened by this work: (i) first, the model can certainly be extended to deal with both incoming users, but also new
items. In that last case, the interview process would consist in asking reviews for any new item to a particular subset of relevant users. (ii) While we have studied the problem of building a static interview - i.e the opinions on a fixed set of items is asked to any new user - we are currently investigating how to produce personalized interviews by using sequential learning models i.e reinforcement learning techniques.