An Effective Model for Jaccard Coefficient to Increase the Performance of Collaborative Filtering

Due to the advancement of technology and an increased number of digital devices per person, more and more digital data are generated daily. Extracting required data from such big data is a challenging task. Recommender systems help us in finding data that best match one's taste. Collaborative filtering (CF) is the most popular approach used in recommender systems. Various similarity measure techniques are used in CF to calculate item-to-item and user-to-user similarity. The majority of these methods use common ratings to compute similarity. One of the similarity measurement methods is Jaccard similarity, which ignores both absolute values of ratings and the average rating value of a user. In this paper, we propose an improved measure that considers the ratio between absolute rating values and number of commonly rated items. We further improved the performance of proposed similarity measure by putting some thresholds on the average rating value of a user. An important aspect of ratings provided by a user is the rating preference behavior of a user, which almost all similarity measurement methods ignore. We also incorporated this behavior in our proposed method. The proposed method is tested over five publicly available datasets: Epinions, FilmTrust, Movie Lens-100K, CiaoDVD and MovieTweetings. The proposed method is compared with various modern similarity measures, and results show improvements in terms of prediction quality and accuracy.