基于矩阵分解和Meanshift聚类的协同过滤推荐算法

Collaborative Filtering Recommendation Algorithm Based on Matrix Decomposition and Meanshift Clustering

摘要: 可扩展性、数据的稀疏性及用户的冷启动问题是传统的协同过滤推荐算法所面临的主要问题。由此提出一种基于矩阵分解和Meanshift聚类的协同过滤推荐算法：首先将原始矩阵使用奇异值分解(SVD)方法进行矩阵分解，较好地对原始数据进行降维，然后使用Meanshift (均值漂移)聚类对所有的物品进行聚类，最后在聚类后的类别中结合改进的基于物品的协同过滤算法，进而减少邻居商品的搜索范围。此方法不仅提高了推荐速度，还良好地解决了用户冷启动问题及数据稀疏问题，在MovieLens 1M数据集上相比于传统的基于物品的协同过滤算法MAE值最多下降了4.52%。

Abstract: Scalability, sparseness of data and cold start of users are the main problems faced by traditional collaborative filtering recommendation algorithms. A collaborative filtering recommendation algorithm based on matrix decomposition and Meanshift clustering was proposed. Firstly, the original matrix was decomposed by singular value decomposition (SVD) method, and the original data would be better reduced. Then Meanshift clustering applied to all items, and finally combined the improved item-based collaborative filtering algorithm in the clustered categories to reduce the search range of neighbors. This method not only improves the recommendation speed, but also solves the user’s cold start problem and data sparse problem properly. Compared with the tradi-tional item-based collaborative filtering algorithm, the MAE value of this method on MovieLens 1M data set is reduced by 4.52%.