1 / 2019-05-06 14:38:27

Maximal Information Based Similarity Measure for Collaborative Filtering

Maximal Information Coefficient, Collaborative Filtering, Similarity Measure

The memory-based collaborative filtering (CF) algorithm has been widespread as it can help users discover their favorite items or products. Finding the similarities among user or items is the critical step for memory-based CF. But, conventional similarity measurement approaches significantly enhanced for collaborative filtering recommendation. We proposed the maximal information coefficient (MIC) which based on mutual information as a measure of similarity to improve the accuracy of CF. We experimented with the benchmark MovieLens-100k dataset and captured item-based CF as an example to demonstrate the strength of our strategy. The empirical results show that our MIC technique accomplishes preferable prediction accuracy over the conventional strategy of measuring similarity.