Research Of Personalized Recommendation For Film Based On Mixed Mode

The trend of information explosion can't be reversed, all those make us in a situation of low utilization of information. So it is more and more important to find a way to search helpful information timely and accurately. Films act as an indispensable part of people's lives, bring us lots of fun. Nowadays the movies'market has the characteristics of rapid expansion and different qualities, so how to offer users a better viewing experience becoming a very important field.In this paper, the author use a hybrid collaborative recommendation algorithm based on feature augmentation, and use both item-based and user-based collaborative filtering technology to double padding the user's movies data matrix, which aims at alleviate the sparsity problem. And using the conversion technology between quantification and quality based on cloud model to measure the similarity among users. This cloud model do not need any user's attributes matching rigorously, actually in this situation you can make full use of the user's score information to calculate the target user's nearest neighbors.In this experiment we use an efficient and stand-alone recommendation engine: Taste, which is belongs to Mahout. Experimental result shows that this model can effectively solve the problem of data sparseness, and improve the quality of recommendation system.The main contribution of this paper:(1) Using cloud model in the framework of Mahout. Compared with the traditional similarity calculation model, the cloud model can obtain more accurate neighbors, and find the feasibility of cloud model with the experiment.(2) Using both item-based and user-based collaborative filtering technology to double padding the user's movies data matrix, to meet user demand for personalized movies. And using cloud model make full use of users'ratings. Implement multiple solutions in dealing with the problem of data sparsity.(3) By using a lot of data, making a detailed test of algorithm in this paper. And using two parameters:Recall and MAE to support the correct of this paper's algorithm and model.