Sparsity Analysis of New Biased Pearson Similarity Measure for Memory Based Collaborative Filtering

In recent time, Collaborative filtering has become the state of art of drive many efficient recommendation systems because of its cross-domain applicability. The core of memory-based collaborative filtering is to compute neighbors for users or items through a similarity measure based on rating information. Sparsity is one of the major challenges of collaborative filtering and has negative effect on the performance. This research work presents a detailed comparative analysis of some traditional similarity measures with the new Biased Pearson Correlation Coefficient (BPCC) similarity measure over a sparse dataset. The experiment was carried out over MovieLens100K dataset with 5 different sparsity levels. The obtained results show that Biased Pearson Correlation Coefficient similarity metric outperformed other traditional similarity measures and gives promising results with an improve of approximate 2% accuracy over RMSE, MAE and MSE evaluation metrics. The similarity measure also performed better over sparse data in terms of prediction accuracy and recommendations classification.