一种融合用户上下文信息和动态预测的协同过滤推荐算法

Collaborative Filtering Recommendation Method Combining User Context Information and Dynamic Prediction

协同过滤推荐技术是应用最广泛、最成功的推荐技术,但面临着数据稀疏性问题和冷启动问题的严峻挑战.同时传统协同过滤算法在相似度计算中忽略了用户个人上下文信息对相似度的影响.提出一种基于用户上下文信息和动态预测的协同过滤算法.首先引入用户上下文信息来改善相似性度量方法,更加真实的反映用户相似度;然后在推荐生成阶段,采用能够充分利用最近邻居集的动态预测方法来进行评分预测.通过在Movie Lens-1M数据集上的实验结果表明:该算法能够缓解评分数据稀疏性对协同过滤推荐算法的影响,显著降低平均绝对误差,提高推荐准确率.

As a most widely used and successful recommendation technology,collaborative filtering recommendation system faces data sparsity and cold start problems. At the same time, Traditional collaborative filtering algorithm do not consider the users context information when calculate the similarity of users. This paper proposes a collaborative filtering recommendation method combining user context information and dynamic prediction. To improve the calculation method of similarity, this paper introduces user context information and rating preference firstly. And then in order to predict the user preference more accurately, this paper proposes a dynamic prediction method which can make full use of the nearest neighbors. Experimental results on MovieLens-1M data sets show that the algorithrn can relieve the influence of rating data sparsity on recommended results and significantly reduce the mean absolute error and effectively improve the recommendation precision.