Soft Cosine Gradient and Gaussian Mixture Joint Probability Recommender System for Online Social Networks

Abstract

Recommender systems are in recent past becoming more notable in the age of swift evolution of the information technology ensuring e-commerce users' suitable items. In recent years, various thread-based and trust based approaches have been proposed, which improve the precision and recall of recommendation to certain extent. However, these approaches are less accurate than expected when users' preferences evolve over time and when several aspects of the users are involved. To solve these problems, we propose a novel recommendation method called, Soft Cosine Gradient and Gaussian Joint Probability (SCG-GJP) for Online Social Networks (OSN). Specifically, the Local Soft Cosine Clique Clustering (LSCCC) is first used to identify good candidate items to be recommended. Second, using adjacency matrix factorization, probable past preference of individual user and overall preference of whole community is formed to reduce the problem of mean absolute error. Again, the obtained matrix is trained through the stochastic function gradient to rank the most suitable candidates. Finally, Gaussian Mixture Joint Probability Postfiltering model is proposed to provide useful recommendations. The experiments are executed with real-world dataset (DS) and compared with the conventional recommendation methods. Experimental results demonstrate that the SCG-GJP obtains higher accuracy, and lesser error as well as complexity.