Collaborative Machine Learning

In information retrieval, feedback provided by individual users is often very sparse. Consequently, machine learning algorithms for automatically retrieving documents or recommending items may not achieve satisfactory levels of accuracy. However, if one views users as members of a larger user community, then it should be possible to leverage similarities between different users to overcome the sparseness problem. The paper proposes a collaborative machine learning framework to exploit inter-user similarities. More specifically, we present a kernel-based learning architecture that generalizes the well-known Support Vector Machine learning approach by enriching content descriptors with inter-user correlations.