论如何利用挖掘社交资讯来改进推荐系统

论文摘要

随着电子商务及移动商务的发展,推荐系统在研究领域和实践领域都显得越来越重要。传统的推荐系统研究者利用用户的打分列表作为推荐的基础,而忽略了能够影响用户偏好的其他因素。然而,实际上用户偏好以及他们的购买决策同时取决于他们自身的经验以及社交信息。以前也有推荐系统研究者在推断用户偏好和打分的时候考虑了社交信息,但是他们只集中在用户的行为（这会带来用户隐私问题）或利用用户之间的距离去计算用户之间的影响力。此项研究着重考察如何充分利用社交信息做更为精准的推荐服务。社交信息指的是从社交环境中获取的信息,不仅包括从亲密的朋友处得来的信息,也包括从大众点评中所获取的信息。根据从社会学,行为学以及推荐系统的研究成果所获得的理论基础,我们提出三种新型的推荐系统方法。在此项研究中的实验比较了现存的推荐方法（包括传统经典推荐算法,近期经典推荐算法以及以前的社交推荐算法）,和此研究中所提出的三种新型推荐算法（包括考虑到从用户朋友网络中挖掘出的社交信息的推荐算法,考虑到从大众点评中挖掘出的社交信息的推荐算法,以及考虑到从用户朋友网络和大众点评中挖掘出的社交信息的推荐算法）的推荐准确性。实验结果表明此项研究中新提出的三项推荐算法能够比之前的推荐算法在准确性,覆盖率以及F-measure等指标中都有更好的表现。特别是当用户没有提供任何打分的情境下,此三种推荐算法是不可取代的,因为其它现存的推荐方法在这种情境下是完全无法工作的。我们也发现从用户朋友网络挖掘社交信息比从大众点评挖掘社交信息更能帮助推荐系统准确的预测用户偏好和用户打分。

论文目录

• 摘要
• Abstract
• 1 Background
• 1.1 Recommender system
• 1.1.1 Recommender system introduction
• 1.1.2 Traditional recommender system
• 1.1.3 Problems existing in traditional recommender system
• 1.1.4 Previous studies to solve these problems
• 1.2 Social network and recommender system
• 1.2.1 Social recommendation introduction
• 1.2.2 Correlation between the Social network and user preference
• 1.2.3 Studies of recommender system combined with social network
• 1.2.4 Correlation between the social network structure and user preference
• 1.3 Public comments and recommender system
• 1.3.1 Problems existing in public comments
• 1.3.2 Correlation between public comments and user preference
• 1.4 Background summary
• 2 Research question
• 2.1 Social recommender system
• 2.1.1 Problems existing in previous social recommender system researches
• 2.1.2 Connection between structure of social network and similarity of preferencesbetween friends
• 2.1.3 How to make use of social information to make rating estimation
• 2.2 Recommender system considering public comments
• 2.2.1 Problems existing in recommender system considering public comments researches
• 2.2.2 How to estimate users' preferences according to users' feedback toward publiccomments
• 2.2.3 Recommender system considering social information mined both from friends andpublic comments
• 2.2.4 Users' preferences leaning
• 2.2.5 Users unknown ratings estimation
• 3 Related works
• 3.1 Holistic review of recommender system researches
• 3.1.1 Content-based recommendation
• 3.1.2 Collaborative-filtering recommendation
• 3.1.3 Hybrid recommender system research
• 3.1.4 User modeling technology in recommender system research
• 3.1.5 Information techniques used in recommender systems
• 3.1.6 Mobile recommender system research
• 3.1.7 Recommendation researches from the perspective of behavior
• 3.2 Existing social recommendation researches
• 3.2.1 Trust-aware recommendations
• 3.2.2 Recommendations combined with social network
• 3.2.3 Researches related to social recommendation
• 3.3 Existing recommender system considering the public comments
• 3.3.1 Word of mouth
• 3.3.2 How online review to influence users attitude towards products
• 3.3.3 The benefits of online review to marketers
• 3.3.4 Usets' implicit feedback
• 3.4 Existing recommender system researches considering social information mined both from friends and public comments
• 3.4.1 How to use public comments in recommender system when combing with socialinformation mined from friends
• 3.4.2 How to use friends' information in recommender system when considering socialinformation mined from public comments
• 3.4.3 Recommender system considering social information mined both from friends andpublic comments
• 3.5 The related works summary
• 4 Theoretical foundations
• 4.1 Social influence
• 4.1.1 Social conformity
• 4.1.2 Social comparison
• 4.1.3 Social facilitation
• 4.2 Social influence and customer behavior
• 4.2.1 Original researches on social influence and customer behavior
• 4.2.2 Social influence and customer behavior
• 4.3 Social network characters
• 4.3.1 Social influence and community
• 4.3.2 Community topology and strength of social influence
• 4.3.3 Centrality and preferences similarity
• 4.4 Users behavior and their preferences
• 4.4.1 Users attitude and users behavior
• 4.4.2 User behavior and implicit rating
• 4.5 Low rank matrix factorization
• 4.5.1 The advantages of matrix factorization
• 4.5.2 Matrix factorization used in recommendation
• 4.6 Theoretical foundations summary
• 5 Recommendation Algorithms
• 5.1 Recommendation considering social information mined from friends
• 5.1.1 Friends V.S.Strangers
• 5.1.2 Friends in one community V.S.Friends in different communities
• 5.1.3 Community size and social influence
• 5.1.4 Community density and social influence
• 5.1.5 Users' centrality and social influence
• 5.2 Recommendation considering social information mined from public comments
• 5.2.1 Products quality and users' ratings
• 5.2.2 Users' unique taste and public comments
• 5.3 Recommendation considering social information both from social network and public comments
• 6 Experiment Design
• 6.1 Tested recommendation methods
• 6.1.1 Previous recommendation methods
• 6.1.2 Our proposed recommendation methods
• 6.2 Tested data set
• 6.3 Tested process
• 6.3.1 Collaborative-filtering recommendation method
• 6.3.2 Matrix factorization recommendation method
• 6.3.3 Previous Social Regularization recommendation method
• 6.3.4 A novel social recommendation method considering social information mined fromfriends
• 6.3.5 A novel social recommendation method considering social information mined frompublic comments
• 6.3.6 A novel social recommendation method considering social information mined bothfrom friends and public comments
• 7 Discussion
• 7.1 Benchmarks for evaluation
• 7.2 Overall performance
• 7.2.1 Collaborative-filtering recommendation method
• 7.2.2 Matrix factorization recommendation method
• 7.2.3 Previous Social regularization recommendation method
• 7.2.4 A novel social recommendation method considering social information mined fromfriends
• 7.2.5 A novel social recommendation method considering social information mined frompublic comments
• 7.2.6 A novel social recommendation method considering social information mined fromboth friends and public comments
• 7.3 New user problem
• 7.3.1 A novel social recommendation method considering social information mined fromfriends
• 7.3.2 A novel social recommendation method considering social information mined frompublic comments
• 7.3.3 A novel social recommendation method considering social information mined fromboth friends and public comments
• 8 Limitations
• 9 Conclusions
• 10 Future works
• References
• Acknowledgements
• 致谢
• 在读期间发表的学术论文与取得的其他研究成果
• 摘要
• 第1章 绪论（推荐系统领域的研究进展）
• 1.1 推荐系统
• 1.1.1 推荐系统介绍
• 1.1.2 传统推荐系统
• 1.1.3 关于解决传统推荐系统所存在问题的研究
• 1.2 社交网络与推荐系统
• 1.2.1 社交推荐介绍
• 1.2.2 社交网络与用户偏好之间的联系
• 1.2.3 关于社交推荐系统的现有工作
• 1.2.4 社交网络结构与用户偏好之间的联系
• 1.3 大众点评和推荐系统
• 第2章 研究问题
• 第3章 文献综述
• 3.1 推荐系统研究的总体回顾
• 3.2 社交推荐系统
• 3.3 利用大众点评改进推荐系统的性能
• 3.4 利用从朋友和在线大众点评中挖掘出的社交信息来改进推荐系统
• 第4章 理论基础
• 4.1 社交影响力
• 4.2 社交网络的结构与社交影响力强度之间的联系
• 4.3 用户行为与用户偏好之间的关系
• 4.4 矩阵分解技术
• 第5章 推荐方法
• 算法1
• 算法2
• 算法3
• 第6章 实验设计
• 6.1 被测试的推荐方法
• 6.2 测试数据
• 6.3 测试过程
• 第7章 实验结果讨论
• 第8章 此研究的局限性
• 第9章 结论
• 第10章 未来的工作
• 参考文献