An Improved Collaborative Filtering Algorithm Based on Bhattacharyya Coefficient and LDA Topic Model

Collaborative filtering (CF) is the most successful method used in designing recommendation systems, which includes the neighbor-based method and the model-based method. Traditional neighbor-based method calculates similarity only based on the rating matrix, but the rating matrix is very sparse. Therefore, to address the problem of sparsity, we proposed an improved collaborative filtering algorithm unified Bhattacharyya coefficient and LDA topic model (UBL-CF). UBL-CF utilized the LDA topic model to mine potential topic information in the tag set and embed the underlying topic information into the progress of the calculation of similarity. Meanwhile, it introduces Bhattacharyya coefficient to alleviate the data sparsity without common ratings. Experimental results show that our method has better prediction in accuracy.