An Improved Collaborative Filtering Algorithm Based on Filling Missing Data

At present, most of the literature uses error metrics to evaluate the performance of recommendation algorithms. However, especially in Top-N recommendation tasks, the accuracy metrics can better show the pros and cons of the recommendation algorithm. Researching traditional recommendation algorithms have a big problem is that the data of recommendation system is always sparse. In order to solve the problem and improve the accuracy metrics, we add a filling matrix when predicting the rating. And then, we considering that different users have different scoring preferences, the user and item biases are added to the loss function. Finally, we use the improved alternating least square method as the optimization method to update the filling matrix in the iterative process. The experiment compared four recommendation algorithms and the results show that, in terms of accuracy metrics, the accuracy of the improved algorithm has been improved many times. In addition, since the filling of the rating matrix needs to consider both the item and the user, we compared the algorithm that considers both and only considers one. The improved algorithm that considered both has an improvement of about 1%.