协同过滤技术及其在电子商务推荐领域的应用研究

【作者】 罗胜阳；

【导师】 哈进兵；

【作者基本信息】 南京理工大学 ， 情报学， 2008， 硕士

【摘要】 随着互联网的普及和电子商务应用的广泛深入,人们在享受网上购物便捷性的同时也陷入了信息过载的困境,用户在大量的产品信息中难以找到自己需要的商品。因此,电子商务推荐系统应运而生。本文对电子商务推荐系统进行了较深入的研究,详细分析了各种个性化推荐技术在电子商务推荐领域的应用现状和前景。在此基础之上,重点研究了电子商务推荐领域的主流技术--协同过滤技术。在详细分析传统的基于项目的(Item-Based)协同过滤算法存在的问题的基础上,提出了一个新的基于项目关联性评分预测的协同过滤算法IAPCF。区别于传统的算法,IAPCF算法不是根据项目之间的相似度,而是根据项目之间的关联规则来寻找项目的最近邻居集合。实验结果表明,IAPCF算法比传统的基于项目的协同过滤算法具有更好的推荐精度。随着电子商务网站规模的增长,基于用户的(User-Based)协同过滤算法存在较严重的数据稀疏性问题,使得用户之间相似度的计算结果不准确,导致推荐质量急剧下降。本文结合上述提出的新算法IAPCF对此进行了改进。改进后的算法IAPCF-UB与传统算法的不同之处在于:在形成用户邻居阶段,计算用户间相似度时,先利用IAPCF算法,根据两两用户间评分项目的并集,预测用户对并集中未评分项目的评分,增加用户之间共同评分的项目数,从而可以解决用户-项目评分矩阵的极端稀疏性,使得计算得到的目标用户的最近邻居更加准确,从而提高算法的推荐精度。实验结果表明,在面对稀疏数据集时,改进算法IAPCF-UB相比于传统的基于用户的协同过滤算法能显著提高推荐质量。更多还原

【Abstract】 With the popularity of the Internet and e-commerce application, consumers enjoy the convenience of shopping on the Internet; on the other hand, they have been in trouble of information overload. It is difficult for them to find their needed products within a mass of product information. Therefore, the recommendation system in e-commerce came into being.In this paper, we made a deep study of recommendation system in e-commerce, and then analysed the status and prospects of the mainstream personalized recommendation techonologies in e-commerce. In this basis, we analysed challenges which collaborative filtering recommendation approach suffered from. And then we proposed an item-association-prediction-based collaborative filtering algoritm (IAPCF) to overcome the shortcomings of the traditional item-based collaborative filtering algorithms. Different from the traditional method, IAPCF algorithm does not use the similarity between items, but the association rules to find the nearest neighbors of target item. The experiment results suggested that IAPCF could provide better recommendation results than the traditional item-based collaborative filtering algorithms.With the expansion of E-Commerce systems, the magnitudes of users and commodities grow rapidly, resulting in the extreme sparsity of user rating data. This situation makes the quality of recommendation systems decreases dramatically. To address this issue, we proposed a collaborative filtering recommendation algorithm based on item rating prediction. The improved algorithm IAPCF -UB we proposed predicted ratings of un-rated item by the IAPCF algorithm, and then the nearest neighbors of target user were calculated with a new similarity measure method. The experiment results suggested that this method could efficiently overcome the extreme sparsity of user rating data and provide better recommendation results than the traditional user-based collaborative algorithms.更多还原