社会网络中兴趣发现与信息组织的研究

【作者】 谢兴；

【导师】 顾宁；

【作者基本信息】 复旦大学 ， 计算机软件与理论， 2011， 博士

【摘要】 在线社会网络(online social network)是一类帮助用户建立人与人之间的在线朋友关系,从而使得人们可以在朋友间分享兴趣和活动的在线服务、平台或Web站点[1]。因为具有丰富的信息自由发布和朋友分享功能,在线社会网络在短短十几年时间中迅速吸引了大量用户的使用,各种类型的在线社会网络应用也是层出不穷。而在产业界取得巨大成功的同时,这些在线社会网络应用也积累了海量用户信息和指数级增长的用户活动数据,从而为学术研究者开展各种分析和应用研究工作提供了宝贵的平台。已经或正在进行的学术研究包括对在线朋友网络结构的分析,对用户活动和其他动态信息的分析和挖掘,以及基于在线社会网络平台的应用研究等多个方面。正是在这些国内外研究的基础上,本文从目前的在线社会网络平台中所存在的信息组织和传播方面的问题出发,在系统架构,朋友关系组织与维护,用户活动和兴趣以及主题信息组织等方面对在线社会网络开展了以下几项研究工作：1)提出了基于层次聚类算法的动态兴趣组构建的在线社区系统结构。本文基于从一个拥有超过63000名用户的在线社区中所收集的超过200万条帖子和1800万条看帖等信息,开展了对在线社会网络平台系统结构,特别是内容组织与用户活动的匹配的分析,发现了目前在线社会网络平台中所存在的灵活的用户兴趣与静态的版块结构之间不匹配的问题。为了解决这种不匹配达到改善在线社会网络内容组织形式的目的,本文提出了基于层次聚类算法的动态兴趣组构建的系统结构,并提供一个灵活的订阅机制,即可同时在内容和用户两个维度上对感兴趣内容进行订阅,从而将用户感兴趣的内容以更高效的方式组织在统一的视图里。2)提出了用户在线会话区间概念和兴趣组内推荐的思想来改善在线社区中内容推送系统的设计。内容推送系统是在线社会网络中一类重要的应用,它使得在线社会网络中的内容传播更具有针对性。本文通过分析什么类型的用户活动更能体现用户兴趣,改进了内容推送系统中用户兴趣的定义方式；通过分析用户在时间上的特定行为模式,帮助确定用户的在线会话区间,从而避免用户兴趣分析中的“假错误”问题；并通过对用户兴趣的倾向性分析,确定了在进行内容推荐时只考虑进行兴趣组内推荐的设计策略。本文基于实际在线社区所设计实施的内容推荐系统也进一步验证了推送效果的改善。3)提出了基于共同兴趣的潜在朋友关系推荐框架,并在特定应用领域设计实现了朋友推荐系统以验证该框架的有效性。本文通过对朋友关系,交互关系以及共同浏览活动之间的异同的分析,提出了基于共同兴趣的潜在朋友关系推荐的思想来改善在线社会网络中内容传播的针对性。此外,针对以往相关研究中推荐准确性不高的问题,提出了基于多维度兴趣融合和结合领域知识的潜在朋友关系推荐框架。为了验证该框架的有效性,本文建立了一个实际的生物领域社区,它包括对小鼠突变品系数据进行发布和分享的数据中心,对生物学家感兴趣的小鼠突变体和基因信息的内容组织,以及基于用户访问数据和基因本体的领域知识所设计实现的帮助生物学家找到具有共同研究兴趣的其他研究者的朋友推荐系统。4)提出了基于在线社会媒体网络的主题信息层次化组织的思想,并设计增量层次聚类算法来保持主题结构模型的实时更新。与现实世界中的社会热点相关的实时讨论信息是在线社会媒体网络中用户感兴趣的一类重要内容。而个人用户往往无法从这些毫无结构化组织且存在大量冗余的内容中获得自己感兴趣主题的完整且精简的结构化描述。针对这样的问题,本文提出了基于在线社会媒体网络的主题信息层次化组织的思想,即将用户感兴趣的社会热点信息组织成以层次化展现不同方面主题的形式并在每个粒度上对主题内容进行概括总结,同时提出了增量层次聚类算法从而能动态地容纳社会热点的新产生信息保持主题结构模型的快速更新。利用搜集于Twitter的350万条社会热点讨论数据,本文验证了层次化组织用户兴趣主题信息的可行性和有效性。更多还原

【Abstract】 Online social network service is an online service, platform, or site that focuses on building social networks or social relations among people to help users sharing interests and/or activities between each other [1]. With the comprehensive functionality of flexibly publishing and sharing information, online social network has enjoyed explosive growth during the recent years and evolved in many different application forms (from online social community to online social media network). In the meantime, the huge success of kinds of online social network applications accumulates massive amount of user profiles and activity data, which further offers a great opportunity for researchers to conduct kinds of data analysis and application designing. These already done or ongoing academic researches include the analysis of online friendship network structure, mining of user activity and other dynamic information in online social network, as well as developing kinds of applications in the platform of online social network.In this work, we focus on the angle of analyzing information organization and diffusion in online social network, and conduct researches on system architecture, friendship organization and maintaining, user activity and interest utilization, and event modeling:1) A novel structure of online social community with the characteristics of dynamic interest group structure is proposed. By conducting a detail analysis on the real and large scale online social community with over 63,000 users,2 million posts, and 18 million views, this paper points out the problem of mismatch between flexible user interest and fixed social group structure. Addressing on this problem, we design a novel structure of online social community with the characteristics of dynamic interest group structure and flexible subscribing mechanism, which can provide dynamic and unique content organization for every user automatically.2) The concept of user session and content recommendation in interest group are proposed to improve the design of content recommendation system. The fast-growing user-generated content poses great challenges on the design of efficient content management and delivery in online social network system. By conducting a comprehensive study of online user activities, including content, social, and time characteristics, this paper try to accurately characterize user interest and user context. And the experiment results obtained in a real online social community further justify that our user interest analysis and context analysis can help to improve the quality and efficiency of content management and delivery.3) A general friend recommendation framework based on common interests and a real friend recommendation system in a biology community are designed. By leveraging interest-based features, we also design a general friend recommendation framework, which can characterize user interest in two dimensions:context (location, time) and content, as well as combining domain knowledge to improve friendship recommending quality. To demonstrate the effectiveness of the proposed framework, the Mutagenesis Information CEnter (MICE) providing biologists publishing, sharing and searching discovered mice mutagenesis information was built. Based on the user accessing requests collected from this data center, we design a friend recommendation system for helping biologists finding other interested researchers.4) User-interested topic information organization solution is proposed. Real-time generated discussion information related to some social events is one of the most important content in online social media network. However, in current online social media network, users are constantly swamped by long streams of unstructured, redundant, and sometimes irrelevant messages, while at the same time lacking a comprehensive and well-organized view of social events. This paper proposes an effective and efficient topic information organization solution (called ETree) with an incremental and hierarchical modeling technique for identifying and constructing event theme structures at different granularities. Detailed evaluation results on event modeling of 20 real social events with 3.5 million tweets demonstrate that ETree can efficiently and incrementally generate high-quality topic structures.更多还原