Grey Forecast model for accurate recommendation in presence of data sparsity and correlation

摘要

Recently, recommender systems have attracted increased attention because of their ability to suggest appropriate choices to users based on intelligent prediction. As one of the most popular recommender system techniques, Collaborative Filtering (CF) achieves efficiency from the similarity measurement of users and items. However, existing similarity measurement methods have reduced accuracy due to problems such as data correlation and data sparsity. To overcome these problems, this paper introduces the Grey Forecast (GF) model for recommender systems. First, the Cosine Distance method is used to compute the similarities between items. Then, we rank the items, which have been rated by an active user, according to their similarities to the target item, which has not yet been rated by the active user; we use the ratings of the first k items to construct a GF model and obtain the required prediction. The advantages of the paper are threefold: first, the proposed method introduces a new prediction model for CF, which, in turn, yields better performance of the model; second, it is able to alleviate the well-known sparsity problem as it requires less data in constructing the model; third, the model will become more effective when strong correlations exist among the data. Extensive experiments are conducted and the results are compared with several CF methods including item based, slope one, and matrix factorization by using two public data sets, namely, MovieLens and EachMovie. The experimental results demonstrate that the proposed algorithm exhibits improvements of over 20% in terms of the mean absolute error (MAE) and root mean square error (RMSE) when compared with the item based method. Moreover, it achieves comparative, or sometimes even better, performance when compared to the matrix factorization methods in terms of accuracy and F-measure metrics, even with small k.