SamenvattingAn approach is described to allow online (streaming) learning of a hidden topic model for use in collaborative filtering recommender systems. The considered topic model is based on Latent Dirichlet Allocation (LDA) . An online, streaming, learning algorithm has two important benefits over batch learning algorithms. First, a streaming algorithm updates the model based on a set of data points, and then never looks at those data points again. This means there is no limit on the size of the total data set, in contrast to batch algorithms that pass through the entire data set in every iteration. Second, an online algorithm enables the quick incorporation of new data, which is important in recommender systems with dynamic user and item sets. The approach described in this document is based on a recently proposed framework for "streaming distributed asynchronous variational Bayesian inference" of a hidden topic model of text documents . Modifications are required to apply the same approach in the setting of collaborative filtering.
|Datum prijs||28 feb 2014|
|Begeleider||Tjalling Tjalkens (Afstudeerdocent 1)|
Online LDA model inference for collaborative filtering
Cox, M. G. H. (Auteur). 28 feb 2014