【作者】 李岩；

【导师】 孔俊；

【作者基本信息】 东北师范大学 ， 计算机应用技术， 2011， 硕士

【摘要】 计算机科学和互联网技术的飞速发展,人们的信息处理能力已经远远落后于信息的产生速度。面对信息过载,搜索引擎只能被动接受指示,无法主动筛选信息。用户如何从海量信息中猎取有价值的信息,或者如何把有价值的信息展示给相关的用户,是近段时间以来学术界热议的一个互联网研究问题。毫无疑问,推荐系统被证明是解决该问题的有效途径。该技术主要是根据分析用户的历史行为信息和兴趣爱好,来达到主动向用户推荐有价值信息的目的。目前大众标注(Social Tag)技术的飞速发展,使得用户可以更加便捷的发现、组织管理或者获取网络上的资源,因而很快就被多个个性化站点所采用。使用标注能够更准确的表述资源特征,也更能真实体现用户个体对资源的兴趣偏好。因此,将大众标注技术用于个性化推荐系统也非常的适用。本文提出了一种基于大众标注聚类的个性化推荐方法,使用归属度(Belonging Coefficient)矩阵代替评分矩阵,不仅可以解决数据稀疏性问题,还可以极大程度上降低数据的维数。与SVD(Singular Value Decomposition)有类似降维思想的推荐算法相比,也在复杂性和推荐效果上都有一定优势。这是对传统推荐方法的改进,可以解决传统方法兴趣模型比较单一的问题。而且这种方法也缩小了评分矩阵的规模,提高了运算的效率。本文以基于MovieLens、Amazon和Netflix等公开的数据集进行的实验表明,基于大众标注的个性化推荐算法与传统的基于用户相似度的方法进行比较分析,可以得出,本文的算法能够显着的提高推荐效果。更多还原

【Abstract】 Information available on internet grows far more rapidly than our ability to process it.Recommender System is one of promising technology to help us find most valuable information without explicit query.It provides recommendations based on user’s history behavior and preferences to entities.This paper puts forward a method based on social tagging clustering method, with Belonging Coefficient matrix instead of score matrix, can not only solve data sparseness, but also can largely reduce data dimension. SVD(Singular Value Decomposition) has similar dimension reduction with the recommended algorithms, the thoughts in complexity and recommend effect has certain advantages. It is recommended to traditional methods of improving the traditional method, which can solve the problem is more onefold interest model. And this method also narrowed score matrix scale, improve the computational efficiency.Based on such MovieLens, Amazon and Netflix data sets based on the experiments show that the mass marked with traditional personalized recommendation algorithm based on user similarity analysis method, this paper, we may conclude that the algorithm can significantly improve recommend effect. Taking the theory of constructivism as the theory base.更多还原