A MODIFIED SPEARMAN’S RANK CORRELATION COEFFICIENT FOR AN EFFICIENT METHOD OF SIMILARITY CALCULATION IN COLLABORATIVE FILTERING-BASED RECOMMENDATION

Author (s) : E. RATHIKA, G. SREEDEVI

---

Abstract :
Collaborative Filtering is among the very popular filtering approaches used widely in e-commerce applications for the recommendation. Based on user similarity, this approach suggests products to the target consumer. The accuracy of collaborative filtering depends heavily on similarity measures. In collaborative filtering-based recommendation systems, low prediction accuracy is caused by erroneous top-n comparable neighbors of the target user. As a result, we have suggested a modified Spearman's Rank Correlation Coefficient in this work, which determines the similarity between users and identifies the target user's similar neighbors. By excluding some ratings from the pre-existing dataset, we purposefully introduce sparsity with variable magnitudes (10 and 20%) in the comparative result analysis of similarity metrics. Following that, the deleted ratings are projected using various similarity criteria. On the Movie Lens dataset, the performance metric MAE is used to compare the proposed technique and the conventional similarity metrics.

Keywords : Recommendation Systems, Correlation, Similarity Metrics, Collaborative Filtering, Prediction Approaches, Nearest Neighbor, MAE.

Conference Date : 13/11/2020

Pages : 344

Paper ID : 20344