A collaborative filtering recommendation algorithm based on correlation and improved weighted prediction

In the case of data sparseness problem, the traditional methods are focused the rating differences of the common rated items instead of the quantity, which causes the unreasonable selection of neighbor set. Moreover, at the stage of predicting ratings, the higher or lower ratings of some individual neighbors will have a great impact on the final prediction. Meanwhile, the choice of mean rating in the Mean Average Method is not scientific because different people may like different kinds of items. To solve the above problems, we first use the cloud model to find users with the closest cloud characteristics to the object user, which can reduce the individual deviation effect. Then we put forward a concept of correlation and find the most the most relevant users and the most relevant items before calculating their similarities, which can make sure there are enough common items between the user and his neighbors and the item and its neighbors. Considering the classification of different items, we find the item neighbors within the relevant items and calculate their mean rating, then we combine the mean item neighbor rating and mean user neighbor rating in an weighted way to predict the final result. Experiments showed that our approach effectively alleviated the data sparseness problem and improved the quality of the recommendation system.