【作者】 王美玲；

【导师】 马军；

【作者基本信息】 山东大学 ， 计算机科学与技术， 2015， 硕士

【摘要】 近年来,互联网不断发展和电子商务普及,为企业发展带来机遇,也向用户提出挑战,如“信息过载”。Web商品信息的爆炸式增长使人们搜索所需信息愈发艰难。在诸多技术中,推荐系统作为个性化服务的一种,能够结合用户具体的背景和兴趣推荐服务,帮助其从浩瀚的网络信息和商品世界中解脱,在电子商务网站中应用广泛。协同过滤则是现有推荐系统中应用最广泛的技术之一。协同过滤针对用户的行为数据分析了解用户的兴趣后再进行推荐。该方法自提出以来就获得普遍关注但自身仍存在一些固有的难题,如数据稀疏、冷启动等。在Web社交网络迅速兴起的形势下,基于社交网络的推荐算法假设用户的兴趣易受其信任朋友的影响,在传统的协同过滤系统中引入了用户之间的社会关系,能有效解决上述难题,逐渐成为目前推荐系统具有研究价值和实践性的课题。在目前流行的社会化推荐算法中,大多数方法都只考虑用户之间的显式信任关系,忽略了隐藏在用户行为下的其他类型的关系。此外,目前已有的信任网络都是二元的,即用户之间有关系,关系值即为1,否则关系值为O。这导致在推荐过程中,当目标用户进行决策时,不论亲疏,其所有信任关系的用户对他的影响程度是相同的,这与现实不符。因为在实际生活中,亲近的朋友影响力往往更大一些。基于上述问题,本文提出了一种结合用户之间加权信任网络与相似关系的社会化模型。围绕如何利用用户之间的关系来进一步提高推荐系统的性能,我们首先对上文提到的二值信任网络进行建模,更精细的计算用户之间的信任程度,然后引入用户的相似关系作为补充,组成用户的邻居集合,最后在进行推荐时,根据用户自身的行为特征来平衡这两种关系的用户对其决策的影响。本文的主要工作和贡献是：第一：总结分析当前个性化推荐系统的研究背景和研究现状,讨论当前个性化推荐系统存在的问题。第二：总结分析个性化服务和推荐系统相关技术,重点介绍协同过滤推荐算法,对其中流行的社会化协同过滤推荐算法的优缺点进行分析。第三：通过对二值信任网络建模,引入用户的相似关系,提出基于加权信任关系和用户相似关系的社会化推荐算法,并在实际数据集Epinions和Ciao上通过实验检验算法预测的准确率。通过对比实验表明,该改进算法与当前流行的社会化推荐算法相比有更高的预测准确率,可以在很大程度上解决数据稀疏、冷启动等问题。更多还原

【Abstract】 Recently, the rapid popularization of Internet and e-commerce not only provide new development chance for the enterprises, but also put forward some challenges, such as "Information overload". Facing exponentially increasing web information, it becomes more and more difficult for people to find out the information they need. As a kind of personalized service, recommender systems can provide different recommendation based on users’ different characteristics to avoid users inundated with choices. Due to their great commercial value, recommendation techniques have been widely adopted in e-commerce.Many techniques have been proposed to make recommendations for users, among which Collaborative Filtering (CF) is one of the most widely used approaches. CF is a technique that can automatically provide recommendations for the target user by collecting information in the form of ratings from other similar users or items. Although it has gained widespread attention, CF has some inherent disadvantages, such as sparsity problem, cold start problem and so on. With the growing popularity of open social networks, approaches incorporating social relationship into recommender systems are gaining momentum. This kind of methods assumes that users are more likely to be influenced by the users they trust in the social network, incorporating the social relation into traditional recommender systems. They can overcome these problems effectively and has become an important research project.Most existing successful social recommendation methods only use the explicit trust relation, which is far from enough. Some other kind of relationships can be exploited, such as the similarity relationship hidden behind users’ behavior. Besides, only few online recommender systems, like Epinions, have the implementation of trust mechanism and the trust values between users are binary. Based on existing binary trust network, these methods assume that different friends trusted by the same user affect the user in the same degree. It is not consistent with the reality.Aiming at solving the problems mentioned above, on the basis of the previous work, we propose a novel recommendation approach based on users’ weighted trust relations and the rating similarities. First, we assign different weights on the social trust relationships among users based on the trustee’s competence and trustworthiness; Secondly, we incorporates the similarity relationships among users as a complement into the social trust relationships to enhance the computation of user’s neighborhood; When generating recommendations, we balance the influence of these two kinds of relationships based on user’s individuality adaptively.The main works in this thesis are as follows:First of all, we analysis the research background and status of personalized recommender systems. We also conclude and analysis the problems of personalized recommender systems.Besides, we conclude and analysis the related technologies of personalized service and recommender systems, especially the Collaborative Filtering. We also analysis the advantages and disadvantages of social recommendation methods, which have been proven very effective.Finally, by modeling the binary trust network and incorporating the similarity relationship, we propose our method and evaluate it on the real datasets Epinions and Ciao for recommendation accuracy. The experiment results show that our proposed method outperforms the state-of-the-art algorithms in improving the prediction accuracy and can alleviate the impact of the sparsity and cold start problems.更多还原