一种基于聚类和协同过滤的组合推荐算法  

A Hybrid Recommendation Algorithm Based on Clustering and Collaborative Filtering

作　　者：刘旭东[1] 葛俊杰[1] 陈德人[2] 

机构地区：[1]烟台职业学院信息工程系,山东烟台264670 [2]浙江大学计算机科学与技术学院,浙江杭州310027

出　　处：《计算机工程与科学》2010年第12期125-127,133,共4页Computer Engineering & Science

基　　金：国家科技支撑计划资助项目(2008BAH21B03)

摘　　要：协同过滤技术是目前电子商务推荐系统中最为主要的技术之一,但随着系统规模的日益扩大,它面临着算法可扩展性和数据稀疏性两大挑战。针对上述问题,本文提出了一种基于聚类和协同过滤的组合推荐算法。首先利用聚类对项目进行分类,在用户感兴趣的类里进行推荐计算,有效地解决了算法的可扩展性问题;接着在每一类中使用基于项目的协同过滤对未评价的项目进行预测,把较好的预测值填充到原用户-项集合中,有效地缓解了数据稀疏性问题;最后根据协同过滤推荐在相似项目的范围内计算邻居用户,给出最终的预测评分并产生推荐。实验结果表明,本算法有效地解决了上述两个问题,提高了推荐系统的推荐质量。Collaborative filtering is one of the main technologies for the e-commerce recommendation systems.However,the lack of algorithm scalability and the sparsity of rating data challenge the gradual increase of users and items.A hybrid recommendation algorithm based on clustering and collaborative filtering is employed to solve these problems.Firstly,the clustering algorithm is utilized to cluster items into several classes.The operations for one user following the clustering algorithm are limited within the interested classes of the user.This strategy improves the scalability of the recommendation algorithm and reduces the computation time.Secondly,an item-based algorithm is employed to compute the predictive values and insert high values into the original matrix in order to relieve the sparsity of the rating data.Finally,a user-based algorithm is used to attain the final predictive value,and then the recommendations are generated.The experimental results indicate that this algorithm can efficiently resolve these problems,and can improve the recommendation quality.

关 键 词：协同过滤 聚类 算法可扩展性 数据稀疏性 平均绝对偏差 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]