【作者】 赵学臣；

【导师】 王洪国；

【作者基本信息】 山东师范大学 ， 计算机软件与理论， 2012， 硕士

【摘要】 随着网络及信息技术的飞速发展，在线购物、社交网络等在线社会网络已成为人们生活中必不可少的一部分。然而网络中的信息呈现出爆炸性增长，信息的数量大大超出了人们的处理能力，“网络信息过载”问题日趋严重。为了满足用户越来越高的信息服务要求，可以向用户推荐感兴趣项目，个性化推荐技术应运而生，并引起了国内外学者和用户的广泛关注。利用个性化推荐技术解决在线社会网络下的信息过载问题，可以有效的提高信息利用率及用户浏览体验，具有重要的现实意义和广阔的应用前景。目前，针对在线社会网络下的个性化推荐问题，国内外研究学者相继提出了一系列的解决方法，为后续的进一步研究奠定了基础。但是现有研究仍存在着以下不足：（1）现有推荐算法普遍只注重用户或资源间的相似度，而忽略了用户兴趣的动态变化，难以快速捕捉到用户兴趣。（2）由于在线社会网络的动态性、不可预知性，传统的网络社团挖掘算法具有低效性且难以反映在线社会网络的动态变化情况。（3）随着推荐系统中的用户和资源数量的不断增加，现有协同过滤推荐算法普遍存在数据稀疏性问题，难以有针对性的选择用户近邻进而给出有效的个性化预测。本文以在线社会网络环境为研究背景，针对当前研究存在的以上问题展开研究，本文的主要工作与创新点如下：1.提一种基于记忆效应的协同过滤推荐算法，与传统推荐算法相比有效的提高了推荐精度，解决了用户兴趣变化下的个性化推荐问题。该模型以认知心理学对人脑记忆-遗忘规律的研究成果为基础，充分分析用户心理，挖掘用户兴趣偏好，将用户的访问行为抽象为人脑的记忆、遗忘过程，使推荐系统中的原始数据随着系统的运行逐渐进行演化，形成不平衡分布，从而在需要时快速抽取出重要数据，分析在记忆、遗忘过程中不断变化的用户兴趣。算法实验表明，引入记忆效应的协同过滤推荐算法，可以更好的反映用户兴趣变化，提高了推荐精度，为现有的推荐系统的设计提供了一种新思路。2.提出一种基于文化的多目标网络社团划分算法，有效克服了动态在线社会网络的不可预知性及传统网络社团划分算法的低效性，解决了动态在线社会网络下的目标团体挖掘问题。该模型对在线社会网络的社团进行着重分析，提出基于文化的多目标社团发现算法，有效的解决了细粒度社团发现问题。该方法不需了解整个网络的拓扑结构便可动态的获得社团的划分数目且可直接得到社团的概貌结构。仿真实验表明，算法在收敛性能及解集分布性方面较传统算法均有改善，并应用于动态网络聚类，取得了良好的聚类效果。3.提出一种基于在线社会网络社团发现的协同过滤推荐算法，克服了传统协同过滤推荐算法推荐生成速度慢及数据稀疏性问题，有效的解决了在线社会网络下的群体推荐问题。针对大数据量环境下传统协同过滤算法的低效性、非实时性及数据稀疏性等问题，将离线网络社团发现及在线推荐相结合，提出基于在线社会网络社团发现的协同过滤推荐算法，使得个性化推荐在同一社团内根据临近用户的喜好相似度进行一致性逼近。更多还原

【Abstract】 As the rapid development of the network and information technology, the online socialnetwork such as online shopping and social network has become an indispensable part of people’slife. But the information of the network shows explosive growth. The amount of information isbeyond people’s handling ability. The problem of network information overload has become moreand more serious. In order to satisfy people’s more and more high information servicerequirements, the interesting projects can be recommended to users. The recommendationtechnology of assisting users’decision arises at the historic moment and calls the domestic andforeign scholars and users’ wide concern. To solve the problem of information overload bypersonalized recommendation technology can effectively improve the utilization rate ofinformation and users’browse experience, which has important practical significance and broadapplication prospects.At present, on account of Personalized Recommendation issues based on online socialnetwork, the domestic and foreign research scholars have proposed a series of the solution, whichlay a foundation for the further research. However, the existing research still has much insufficientas follows:(1) The existing recommendation algorithms only pay attention to the similarity betweenusers and resources, however, they ignore the dynamic changes of users’interests, so it is difficultto quickly capture the user interest(2)As the online social network is dynamic and unpredictability, traditional networkcommunity mining algorithms are inefficient, and it is difficult to reflect the dynamic changes ofthe online social network.(3) With the uninterrupted increase of users and resource, it is common that existingcollaborative filtering algorithms have the problem of data sparseness. We can’t choice users’nearneighbors’nichetargeting and put forward effective personalized forecast.In this paper, the online social network environment is the research background. The paperstudies based on the above-mentioned existing problems. The main work and innovation pointsare as follows:1. In view of the changes of users’interest, it puts forward a collaborative filtering algorithmbased on memory. Compared with the traditional collaborative filtering algorithms, it effectivelyimproves the precision of the recommendation, and it provides a new idea for existing design ofthe recommend system.The algorithm, based on the research achievement of cognitive psychology to the brainmemory-forget rule, fully analyzes users’psychological, excavates users’interest preference,abstracts the user access behavior to the human brain’s memory-forgotten process, makes theoriginal data of recommendation system evolve with the operation of the system and formdisequilibrium distribution, then extracts important data quickly when needed, analyzes thechanging user’s interests in the process of memory and forgotten. Experimental results show thatcollaborative filter recommendation algorithm with memory effects can reflect the change ofusers’interests commendably. The algorithm improves the precision of the recommendation and provides a new idea for existing design of the recommendation system2. In view of unpredictability of online social network and inefficiency of traditional networkcommunity division, the paper gives the optimization problem description of the networkcommunity division and puts forward a multi-objective network community partition algorithmbased on culture, which effectively solve the problem of fine grained community finding.The paper analyzes the online social network community. It Introduces the reality and theorysignificance that the online social network community have for personalized recommendation, andthen puts forward a multi-objective community finding algorithm based on culture, whicheffectively solved the fine grain community finding problem. With this method we need not toknow the topology of the whole network, we can get the division of the club number dynamicallyand can get the community’s management structure directly. The simulation results show that thealgorithm is better in the aspects of convergence and solution set distribution than traditionalalgorithms. It is applied to dynamic network clustering and achieves good clustering effect.3. Aimed at the traditional collaborative filtering recommendation algorithms’problem ofslow generating speed and data sparseness, this paper puts forward a collaborative filterrecommendation algorithm based on online social network community finding. The algorithmsolves the speed problem of generating recommendation to a certain degree, and it improves thequality of the recommendation.In view of the problems of low efficiency、non-timeliness and data sparse in traditionalcollaborative filtering algorithms under the environment of large amount of data, It combinesoff-line network community finding and online recommendation, and proposes a collaborativefiltering recommendation algorithm based on online network community finding, which makespersonalized recommendation in the same community do consistency approximation according tothe similarity degree of near users’interests.更多还原