Decoupled Metric Learning Model combining Tag Information for Recommendation

摘要

Metric learning produces distance functions to capture the important relationships among data and has been explored in collaborative recommendations successfully. Existing metric learning recommendation model cannot express the rating preference of users well, and the differences between items will also have a non-negligible influence on the trained metric space. For this, we present a decoupled metric learning model combining tags information (DMLT) for recommendations. DMLT model does well in decoupling the preference of users from rating data with tags information. Experiments on public datasets show that DMLT model is more accurate than the matrix factorization algorithm and other latest metric learning recommendation algorithms.