基于数据双重优化聚类的协同过滤推荐算法

Collaborative filtering recommendation algorithm based on double optimization clustering with data

在传统的协同过滤推荐算法使用中,不可避免地遇到数据稀疏性、冷启动和可扩展性问题。然而个性化推荐系统需要真实大量的流式数据支持,因此需要一类能够有效提高符合现实用户需求的推荐方法。该类推荐方法必须密切联系用户日常操作行为,采集用户偏好程度等等。在流式大数据环境下,推荐算法中的用户特征和用户偏好尤为重要。文中提出了一种基于用户活跃度(CF-act)双重聚类的协同过滤推荐算法,以用户特征和用户偏好为种子数据,对种子数据采用优化的K-means算法分别聚类生成用户-项目评分表,在聚类中搜索目标用户的最近邻居为目标用户产生推荐,缩小搜索范围,降低推荐的时间复杂度。最后,文中比较了提出的推荐算法与传统协同过滤算法的误差率,且验证了推荐系统的运行效率。

In the use of traditional collaborative filtering recommendation algorithm,data sparsity,cold start and scalability are unavoidable. However,personalized recommendation system requires a lot of real data support,so a kind of recommendation method can be effectively improved to meet the needs of real users. This kind of recommendation must be closely related to the user＇s daily operation behavior,to collect user preference and so on. In the large data environment,the user characteristics and user preferences of the proposed algorithm are particularly important. This paper puts forward a kind of based on user activity（ CF-act） double clustering collaborative filtering recommendation algorithm,characteristics and user preferences to the user data for seed, the seed data, respectively, the optimization of K-means algorithm clustering generate user-project assessment,in clustering search target user nearest neighbors of target user recommendation,narrow your search,reduce the time complexity of recommendation. Finally,it compares the error rate of proposed algorithm and traditional collaborative filtering algorithm,and verifies the operating efficiency of recommendation system.