Abstract:

Till now, collaborative filtering has been the most successful and widely used technology in recommender systems. However, the rating data is extremely sparse so as to affect the prediction accuracy seriously in traditional collaborative filtering. In order to overcome the drawback, in this paper, we proposed three data filling approaches and two recommendation strategies. These data filling approaches for non-rating data in rating matrix are: (1) Filling data using weighed average of row and column data; (2) Filling data using mode average of row and column data; and (3) Filling data using median average of row and column data. One of the recommendation strategies is taking filling data for predicative rating directly, and another is to set the rating matrix filled data as a pseudo rating matrix and collaborative filtering applying Pearson correlation. The experimental results on the MovieLens data set show that all these recommendation strategies can effectively alleviate the trouble of rating data sparseness and gain better recommendation accuracy.

Key words: recommender system;personalized recommendation;collaborative filtering;data filling