基于杰卡德相似性的推荐系统研究

协同过滤推荐系统的核心是用户的相似性度量.在杰卡德相似性度量基础上,提出一种修正的杰卡德相似性度量.该方法将用户评分差异的数目融入相似度计算,并综合杰卡德相似度建立神经网络学习模型,选取Movielens数据作为训练集,得到合适的权重.实验结果表明,与pearson相似性度量相比,该方法在用户评价较少时给出相对可靠的推荐,在推荐的精度、平均绝对误差等方面具有一定的优越性.

Research on recommender systems based on jaccard similarity

The core of collaborate filter recommender systems is user similarity measure. Based on the Jaccard similarity measure, this paper proposes a modified Jaceard similarity measure which joins the difference of ratings between two users, uses optimization based on neural network learning model. In order to get the suitable weights,this metric has been tested on the Movielens databases. The experimental result shows that, to compare with pearson similarity measure, this method can be used to give recommendation to on the condition of the less rating users;and it is superior in the respects of precision, mean absolute error and so on.