A novel approach to solve the sparsity problem in collaborative filtering

摘要

Collaborative Filtering (CF) is the most successful approach of Recommender System. Although it has made significant progress over the last decade, the current CF method is stressed by the sparsity problem. In this paper we propose a novel approach to address this issue. Multiple Imputation (MI) is a useful statistic strategy for dealing with data sets with missing values and replace each missing value with a set of plausible values that represent the uncertainty about the right value. In our approach we apply MI technique in the data processing procedure to turn the original sparse data into dense data. And then we use the dense data and the original data in the following CF progress separately. We compare their performance both in cosine-based and correlation-based similarity measures. We conduct a 10-fold cross validation and take the MAE as the evaluation metrics. Our experimental results show that our approach can efficiently solve the extreme sparsity problem, and provide better recommendation results than traditional CF method.