An Item-Diversity-Based Collaborative Filtering Algorithm to Improve the Accuracy of Recommender System

Abstract

Matrix factorization is a model-based collaborative filtering recommendation algorithm. From the performance of the recommender system of large websites such as Amazon and Facebook in recent years, it can be seen that collaborative filtering technology has achieved important results in theoretical and practical. Although personalized recommendation based on collaborative filtering has been widely used in the world, it still faces some challenges, such as data sparseness, cold start, recommendation accuracy, and novelty issues. This paper proposes an improved recommendation method based on item-diversity, adding bias and implicit feedback. Through research, we find that users prefer multiple choices. To achieve that, we add the term based on variance to the matrix factorization algorithm. It significantly improves the diversity of recommendations. And accuracy also be improved to some extent with diversity growing. By combining the bias and implicit feedback in the improved algorithm, we derive the list of recommendations that users want. The experiment used real data sets. The result show that our optimized model can effectively improve recommendation accuracy and quality.