@techreport{
    Puolamaki04,
    author = {Puolam{\"a}ki, Kai and Savia, Eerika and Sinkkonen, Janne and Kaski, Samuel},
    title = "Two-Way Latent Grouping Model for User Preference Prediction",
    abstract = "We introduce a novel latent grouping model for predicting the relevance of a given document (or item) to a certain user. Our model assumes both users and documents to have a latent group structure. We compare the model against a state-of-the-art method, the User Rating Profile model (URP) where only users have a latent group structure. Both models are estimated by Gibbs sampling. It turns out that our method gives more accurate relevance predictions than the URP model when the task is to predict ratings for comparatively new documents for which only very few ratings are previously known. In such a situation generalization over documents becomes necessary and two-way grouping pays off. This work has been done as a part of a proactive information retrieval project that aims at estimating relevance of new documents to users based on both explicit and implicit user feedback.",
    address = "Espoo, Finland",
    number = "A80",
    flags = "copy public",
    year = "2004",
    type = "Publications in Computer and Information Science",
    institution = "Helsinki University of Technology"
}