摘要
协作过滤算法作为最成功的个性化推荐技术已经被应用到很多领域中。针对现有推荐算法存在的预测值判定不准确、数据高维稀疏性、可扩展性不强的问题,提出了一种改进的协作过滤方法。协作过滤方法首先通过对资源进行分类、加权过滤数据预处理以及K-平均聚类算法对用户进行聚类,然后利用余弦相似度计算用户间的相似性,产生最近邻居集,最后基于可信度对算法产生的预测值进行修正,从而得到最终的推荐集。实验结果表明,改进后的协作过滤算法在推荐效果方面得到了更好的改善。
Collaborative filtering is the most successful personalized recommendation technology, and is extensively used in many fields, the existing recommending approaches have many defects, for example, the inaccurately predict value, the high - dimensional sparse data, the lower scalability. To solve this problem, this paper proposes an improved collaborative filtering algorithm. Firstly, classifying of resources, pre - processing of weighting filter data and clustering the users based on the k - mean cluster algorithm, then, the nearest neighbor set is produced by computing the similarity of users based on Cosine Similarity, finally, the method produces the results by amending the predict value based on the credibility. Experiments result shows that our proposed algorithm outperforms traditional collaborative filtering algorithm.
出处
《电气自动化》
2011年第5期15-17,共3页
Electrical Automation
关键词
协作过滤
聚类
最近邻居集
可信度
预测值
Collaborative filtering Clustering The nearest neighbor set Credibility Predict value