Global e-commerce sites use recommender systems to provide customers products who might be interested in, which aims to increase the sales opportunity, but if the recommendations are similar, so customers lose interest in the recommendations, indirect effects of customer loyalty and e-commerce website to sell more products. recommender system which provides diversity is the main method to solve this problem, the objectives of diversity means to cover the customer widespread interest and to provide more dissimilar recommendations.
In this paper, we present a collaborative filtering approach to improve the diversity of the individual and aggregation, which is to place diversity of factors into the "for-mation of neighborhood" step and "produce recommended list" step. We use dataset from well known MovieLens to evaluate our model. We use hybrid weight method in the "to form a neighbor" step to find the most k neighbors of high diversity and then obtain the predicted rating value of every single movie items of the target users by the k neighbors. Then ranks movie items directly based on their predicted rating value from highest to lowest. We use Item Popularity、Item Average Rating and Item Absolute Likeability ranking method to re-rank the movie items.Finally we get a high diversity movie items list.
Six Measures(precision, recall, F1-measure, individual diversity, aggregation di-versity and relative benefit) are used to evaluate the performance of the system. The experimental results that placing diversity of factors into the "formation of neighbor-hood" step can improve individual diversity and aggregation diversity, but placing di-versity of factors into the " produce recommended list " step can only improve aggrega-tion diversity.
