基于近邻传播聚类的混合推荐系统

协同过滤技术是推荐系统最具价值的核心技术之一，它能够深入地挖掘用户潜在的兴趣爱好并向用户做出比较合理的推荐；但是冷启动、数据稀疏性、可扩展性等问题依然制约该技术在实际推荐系统的应用。针对冷启动和数据稀疏性等问题，文章提出了一个基于近邻传播聚类的混合协同过滤推荐模型。该模型首先基于物品的标签属性进行聚类，挖掘出同类的物品并计算相似物品之间的关联程度，然后基于历史交互数据计算物品的相似度矩阵，最后按照一定权重混合构成一个物品相似度，并以此为用户进行推荐。与传统协同过滤推荐模型相比，该模型不仅提高了推荐精确度，而且改善了物品的召回率，能为用户提供更好的推荐体验。

Hybrid Recommendation System Based on Affinity Propagation Clustering

Collaborative filtering(CF) is one of the most valuable technologies of recommendation system. It can effectively mine users' potential hobbies and make reasonable recommendations to users. But the technology applied in the actual recommendation system is still constrained by cold start, data sparsity, scalability and so on. In this paper, a hybrid collaborative filtering recommendation model based on affinity propagation clustering(AP) was proposed for cold start and data sparseness. The model first clustered based on the tag attributes of items, and mined items of the same type and calculated the degrees of association between similar items. Then, the historical interaction data was used to calculate the similarities of all items in the model. Finally, a certain proportion was mixed to form an item similarity matrix which was used to recommendation for users. Compared with the traditional collaborative filtering recommendation model, it not only greatly improves the recommendation precision, but also improves the recall rate of the item and provides a better recommendation experience for the user.