【作者】 李林；

【导师】 杨磊； 杨颖；

【作者基本信息】 广西大学 ， 计算机应用技术， 2016， 硕士

【摘要】 基于位置的服务(Location-Based Service, LBS)和社交网络逐渐融合,形成了基于位置的社交网络(Location-Based Social Networks, LBSNs)。而随着用户与签到的数量不断剧增,将具有信息过滤能力的推荐系统引入到LBSNs中,能够较好地帮助用户缩短寻找真正关注内容的时间,提高获取需求的效率。本文主要研究LBSNs推荐技术中的潜在好友推荐系统。在对LBSNs网络层次结构,常用相似性计算方法以及几种好友推荐算法进行研究,分析和对比他们的优势和不足的基础上,提出一种基于社交关系和签到行为的潜在好友推荐算法,并将其运用到原型系统的设计与实现中,以便提高好友推荐的效果。具体工作如下：针对现有的LBSNs好友推荐算法没有很好地对社交关系进行分析的问题,提出一种基于社交关系和签到行为的潜在好友推荐算法PFRSC。在社交关系的考虑上,先根据目标用户与邻居节点之间的共同好友数求出直接关系值,再利用关系的传递性计算出与目标用户在社交关系上的待推荐用户集合,更好地表示了关系的强弱。在计算用户间签到行为相似性上,提出一种通过签到频率和签到比例来对签到次数进行归一化处理的新方法,综合考虑了用户个人偏好和大众偏好,解决了传统算法中只能考虑共同签到的个数,而无法考虑次数等问题,较好地提高发现潜在好友的效率。最后,以准确率和召回率作为潜在好友推荐效果的度量,通过实验证明了所提出的PFRSC算法比传统的好友推荐算法具有更好的推荐效果。以PFRSC算法为基础,通过对好友推荐系统进行调研和需求分析,设计并实现了包括总体结构、功能模块以及数据库等方面的基于位置的潜在好友推荐系统原型。它能为当前用户预测潜在好友,并按照社交关系和签到行为的综合相似性大小进行重排序,最终以推荐列表的形式提供给用户,为用户选择提供依据,有利于帮助用户建立和扩展自己的社交圈子。更多还原

【Abstract】 While the LBS(Location-Based Service) and social networks gradually mix together, LBSNs (Location-Based Social Networks) has emerged. And with the number of users and check-ins in LBSNs continue to increase, the introduction of recommender systems which has the ability of information filtering in LBSNs will better help users shorten the time to find the contents they really interested in, and improve the efficiency of getting their requirement.This paper mainly research the potential friends recommender systems which is one of the aspects in LBSNs recommendation technology. Through the study of LBSNs’hierarchical network structure, similarity calculating methods and some friend recommender algorithms, comparing their merit and demerit, this paper presented a new friends recommender algorithm based on social relationship and check-in behavior, and use it in a prototype design and implement of a potential friend recommender system in order to promote the effect of recommending potential friends. The specific research works are as follows:Aimed at the problem that traditional friend recommender algorithms in LBSNs do not have a better way to analyze social relationship, this paper presented a new friends recommender algorithm based on social relationship and check-in behavior called PFRSC algorithm. In the consideration of social relationship, according to the common friends of the target user and its neighbor node, PFRSC first computes the direct relationship value of each other, and then using the relationship transitivity to calculate the similar user set in social relationship for target user, it can better represent the intensity of social relationship.When computing the check-in behavior similarity between users, this paper presented a new approach based on check-in frequency and check-in ratio, which normalization process for common check-in times, comprehensively considerate user’s personal preference and public preference to calculate the check-in behavior similarity, it can solve the problems in traditional algorithms such as only considering the number of common check-in locations without considering the common check-in times, better improve the efficiency of finding potential friends. Last, using the precision and recall as the measurement approach for recommending effect, the experiment shows that PFRSC has a better recommending effect than the traditional friends recommending algorithms.Based on the PFRSC algorithm, through the investigation and requirement analysis of potential friends recommender system, this paper has designed and implemented a potential friends recommender system prototype which including architectural structure, functional module and database design etc. The system can predict the potential friends for the current user, ranking the comprehensive similarity of social relationship and check-in behavior from high score to low score, finally shows a friend recommender list to the user, providing a reasonable basis to make a choice, helping the user to establish and expand their social circle.更多还原