基于网络分割聚类的标签语义规范化推荐算法

传统的推荐算法多以用户评分数据计算用户的兴趣偏好以及资源相似度,对稀疏数据以及新用户的推荐质量较低。考虑到用户标签数据的随意性和语义模糊性,提出基于标签网络分割聚类的语义规范化方法并建立基于规范化标签的用户兴趣模型。该模型能在不改变用户兴趣的前提下有效降低用户标签兴趣模型的向量维数,并能避免分析标签语义的复杂过程,且能根据用户自身的理解来获取用户兴趣。最后将标签兴趣模型应用到推荐算法中。通过与经典的推荐算法进行比较,验证了该算法能有效缓解数据稀疏性、推荐冷启动问题,提升了推荐结果的准确性,能获得更好的推荐效果。     

基于图嵌入模型的协同过滤推荐算法

传统协同过滤算法存在严重的数据稀疏和冷启动问题。利用社交网络中的丰富信息为解决传统协同过滤算法的数据稀疏和冷启动带来了契机。然而,传统基于社交网络的协同过滤算法仅利用粗粒度、稀疏的用户信任关系来改进传统协同过滤算法,即用0或1表示用户之间信任程度。另外,传统基于社交网络推荐算法仅仅集成用户之间显式信任关系,而忽略用户之间隐式的信任关系。本文提出一种基于图嵌入模型的协同过滤推荐算法,即利用图嵌入模型技术学习社交网络中用户的低维特征表示,并根据用户的低维特征表示推导用户之间细粒度的信任关系。最后,根据信任用户和相似用户对目标物品的评分权重预测用户对目标物品的评分。在真实数据集上的实验结果表明,基于图嵌入模型的协同过滤算法的性能优于传统的协同过滤算法。     

基于用户信任和张量分解的社会网络推荐

社会化网络中的推荐系统可以在浩瀚的数据海洋中给用户推荐相关的信息。社会网络中用户之间的信任关系已经被用于推荐算法中,但是目前的基于信任的推荐算法都是单一的信任模型。提出了一种基于主题的张量分解的用户信任推荐算法,用来挖掘用户在不同的物品选取的时候对不同朋友的信任程度。由于社交网络更新速度快,鉴于目前的基于信任算法大都是静态算法,提出了一种增量更新的张量分解算法用于用户信任的推荐算法。实验结果表明：所提出的基于主题的用户信任推荐算法比现有算法具有更好的准确性,并且增量更新的推荐算法可以大幅度提高推荐算法在训练数据增加后的模型训练效率,适合更新速度快的社会化网络中的推荐任务。     

基于时空数据的社会关系发现*

网络技术的发展产生了大量的网络用户,他们之间潜藏的社会关系越来越多地引起了人们的注意,大量的社交网络发现算法已被提出。但是,以前的研究多建立在关系数据可直接获取的基础之上。实际上,网络数据多以用户个体行为形式存在,数据实时变化。基于用户使用网络的行为日志分析,提出基于时空数据分析模型的社会关系发现算法,算法主要包括实际分析和关系发现两个步骤。通过实验表明,本算法能很好地发现用户行为中潜藏的社会关系。     

基于SRank的社交网络影响力分析

针对社交网络中用户影响力的评价问题,提出了一种基于SRank的评价算法。基于从社交网络中收集的大规模数据集,结合最近社会学理论研究成果分析PageRank及其改进算法应用于此场景中的不足。在此基础上总结社交网络中信息传播的规律,将用户与社交网络的关系强度定义为用户的人缘值,用来表示用户作为粉丝的信息再传播能力。然后提出了一个通过预测用户信息传播能力大小来分析和度量用户影响力的SRank用户影响力模型。在同样的数据集下相对于PageRank及其改进算法,SRank用户影响力模型获得了更好的影响力预测结果。基于大规模数据的实验结果表明,提出的方法是较为有效的。     

基于深度神经网络的社会媒体网络分析

社会媒体网络中不仅包含了用户、文本、图片和视频等多种模态的数据,还包含了反映不同模态数据之间交互的群体特征。为了更好地描述社会媒体网络,从而为上层应用提供更好的服务,提出了一种基于深度神经网络的社会媒体网络模型。该模型采用深度神经网络对单个模态的数据进行学习,从而得到任意一个模态数据的潜在特征表示方法。对于两种不同模态的数据,利用具有高斯分布的先验矩阵与两个模态数据的后验分布建立反映这两个模态数据间群体特征的生成模型。实验结果表明,提出的模型在网络结构的链接分析中具有更好的预测效果,能有效地描述社会媒体网络的整体特征。     

基于贝叶斯网络的现货市场化交易用户电能量预测

为提升电能量预测准确性,研究基于贝叶斯网络的现货市场化交易用户电能量预测方法。利用主成分分析降维处理用户电能量数据;以余弦相似度为降维后数据聚类的形态度量相似判据,建立用户电能量特征库;利用评分搜索与期望最大化算法确定贝叶斯网络结构与学习网络参数,建立贝叶斯网络模型,在该模型内输入特征库内的数据,输出电能量预测结果。实验证明：该方法可精准聚类用户电能量数据曲线,精准预测用户电能量;该方法网络参数学习时的相对熵距离较小,具备较优的网络参数学习效果。     

基于共同用户和相似标签的好友推荐方法

针对目前的社交网络好友推荐方法用户兴趣不明显、用户之间相关性较差等问题,提出一种基于共同用户和相似标签的协同过滤算法。抽取共同关注用户作为共同项目,加入体现用户兴趣的自定义标签数据,并对标签进行相似度计算处理,以扩充稀疏矩阵,改善协同过滤推荐方法。实验结果表明,与单指标的协同过滤推荐算法相比,基于共同用户和相似标签的好友推荐方法更好地体现了用户兴趣,同时在推荐准确率和平均准确率上都有较大提高。     

基于用户偏好的网络质量QoE感知建模仿真分析

为了解决现有网络质量QoE感知模型数据粗差迭代次数多、线性回归参数小的问题,提出基于用户偏好的网络质量QoE感知建模仿真研究。依据用户偏好理论确定模型参数,并获取网络质量QoE感知数据,以此为基础,通过MCD算法判别并去除网络质量QoE数据粗差,以去除粗差的网络质量QoE数据为基础,利用ROI加权算法提取网络质量QoE数据特征,以得到的网络质量QoE数据特征为依据,将其代入多元线性回归方程计算网络质量QoE感知,实现了基于用户偏好的网络质量QoE感知。实验结果显示,与现有三种网络质量QoE感知模型相比较,构建的网络质量QoE感知模型降低了数据粗差迭代次数,提高了线性回归参数,充分说明构建的网络质量QoE感知模型具备更好的性能。     

位置社交网络中基于评论文本的兴趣点推荐

随着位置社交网络(Location-Based Social Networks,LBSN)的快速发展,兴趣点(Point-Of-Interest,POI)推荐对于用户和商家愈发重要。目前基于社交网络的兴趣点推荐算法主要利用用户的历史签到数据和社交网络数据来提升推荐质量,但忽略了利用兴趣点的评论文本数据;并且LBSN中的数据经常会存在部分信息缺失的情况, 对兴趣点推荐算法而言如何保证鲁棒性是一个巨大的挑战。为此,提出了一种新的用户兴趣点推荐模型,称其为SoGeoCom模型。该模型融合了用户社交网络数据、地理位置数据以及兴趣点的评论文本数据这3个因素来进行兴趣点推荐。基于来自Yelp的真实数据集的实验结果表明,与其他主流的兴趣点推荐算法相比,SoGeoCom模型能够提高准确率和召回率,并且具有良好的鲁棒性,获得了更好的推荐效果。     

Aligning principal and agent’s incentives: A principal–agent perspective of social networking sites

The viability of networked communities depends on the creation and disclosure of user-generated content and the frequency of user visitation (Facebook 10-K Annual Report, 2012). However, little is known about how to align the interests of user and social networking sites. In this study, we draw upon the principal-agent perspective to extend Pavlou et al.’s uncertainty mitigation model of online exchange relationships (2007) and propose an empirically tested model for aligning the incentives of the principal (user) and the agent (service provider). As suggested by Pavlou et al., we incorporated a multi-dimensional measure of trust: trust of provider and trust of members. The proposed model is empirically tested with survey data from 305 adults aged 20-55. The results support our model, delineating how real individuals with bounded rationality actually make decision about information disclosure under uncertainty in the social networking site context. There is show little to no relationship between online privacy concerns and information disclosure on online social network sites. Perceived benefits provide the linkage between the incentives of principal (user) and agent (provider) while usage intensity demonstrated the most significant impact on information disclosure. We argue that the phenomenon may be explained through Communication Privacy Management Theory. The present study enhances our understanding of agency theory and human judgment theory in the context of social media. Practical implications for understanding and facilitating online social exchange relationships are also discussed.     

The Network Awareness Tool: A web 2.0 tool to visualize informal networked learning in organizations

Professionals often initiate informal learning in an effort to solve work-related problems. This paper focuses mainly on research methods for visualizing informal networked learning for teachers. Drawing upon Social Network Theory and Analysis and such notions as Social Capital, Networked Learning, and Communities of Practice we create a theoretical model to underpin the design of our web 2.0, Network Awareness Tool NAT.NAT address several problems connected with informal networked learning research in organizations, such as the informal learning paradox. Advanced Search and Signalling Features can overcome the problem of under-representation of the data. Authorship minimises the problem of unrealistic representation of data. Making use of user profiles and allowing them to (re)enter and improve data on both online and off-line (and combinations thereof) networked activities offers an alternative approach to the name/resource generating (Van der Gaag & Snijders, 2005) and position generating strategies (Lin, Fu, & Hsung, 2001) often used for collecting social network data. The implementation of a profile page, tags and ratings provides the opportunity to triangulate data on the needed multidimensional level, representing informal learning in the dimensions covered by the theoretical model. NAT could impact on our current understanding of informal networked learning.     

基于作者主题模型和辐射模型的用户位置预测模型

由于全球定位系统（GPS）设备采集的用户历史位置数据通常具有稀疏性,使得基于单个用户数据的位置预测模型能力受限,所以结合人类移动性的时间和空间周期性,提出一种基于作者主题模型（ATM）和辐射模型（RM）的用户位置预测模型。在时间维度上,该模型利用ATM发现与目标用户移动行为相似的用户群,并确定该用户群在预测时刻所处的目标状态;在空间维度上,该模型利用RM算法计算目标用户的候选地点在目标状态下的概率,并通过比较各候选地点的概率值确定目标用户可能出现的地点,从而实现对目标用户位置的预测。实验结果表明,该模型的平均预测准确率为61.49%,相对于基于变阶的Markov模型提高近28个百分点。所提预测模型能够在单个用户数据量小的条件下获得更高的预测准确率。     

基于GNN的矩阵分解推荐算法

相较于协同过滤,矩阵分解有着更好的拓展性和灵活性,但同样受到数据稀疏和冷启动的困扰。针对上述问题,提出一种融合GNN和PMF的推荐算法GNN_MF。该算法通过神经网络对社交网络图以及用户项目图进行建模,将两个图内在的联系起来,学习目标用户在社会空间以及项目空间上的特征向量。通过MLP将两个特征向量串联提取用户的潜在特征向量,集成在概率矩阵分解模型上,产生预测评分。在真实数据集Epinions、Ciao上的大量实验表明,GNN_MF算法的均方根误差和平均绝对误差较传统PMF分别降低了2.91%、3.10%和4.83%、3.84%。验证了GNN_MF算法在推荐系统中的有效性以及可行性。     

融合社会标签与信任关系的社会网络推荐方法

针对推荐系统中普遍存在的数据稀疏和冷启动等问题,本文将标签与基于信任的社交推荐方法相结合,提出了一种融合社会标签和信任关系的社会网络推荐方法。该方法利用概率因式分解技术实现了社会信任关系、项目标记信息和用户项目评分矩阵的集成。从不同维度出发,实现了用户和项目潜在特性空间的互连。在此基础上,通过概率矩阵因式分解技术实现降维,从而实现了有效的社会化推荐。在Epinions和Movielens数据集上的实验结果表明本文所提出的方法优于传统的社会化推荐和社会标签推荐算法,特别是当用户评分数据较少时该算法的优越性体现得更好。     

社交网络中融合社交关系和语义信息的推荐算法

协同过滤方法广泛应用于推荐,但是数据稀疏成为模型提供高质量推荐的一大障碍.为了解决此问题,文中提出融合社交关系和语义信息的推荐算法,提高协同过滤方法的推荐性能,有机融合稀疏的用户行为记录、项目的社交信息和项目的语义信息.应用矩阵分解技术把行为矩阵和项目社交关系映射到一个低维的特征空间,提供项目社交关系信息分解的显式解释,分析关系信息对用户行为偏好产生的影响.同时,使用社会化因子正则的级联去噪自编码器模型学习项目语义特征,改进传统深度学习模型.在真实腾讯微博和Twitter数据集上的实验表明,文中方法有效提高召回率、准确率和推荐效率.     

A semantic and social-based collaborative recommendation of friends in social networks

The development of social media technologies has greatly enhanced social interactions. The proliferation of social platforms has generated massive amounts of data and a considerable number of persons join these platforms every day. Therefore, one of the current issues is to facilitate the search for the most appropriate friends for a given user. We focus in this article on the recommendation of users in social networks. We propose a novel approach which combines a user-based collaborative filtering (CF) algorithm with semantic and social recommendations. The semantic dimension suggests the close friends based on the calculation of the similarity between the active user and his friends. The social dimension is based on some social-behavior metrics such as friendship and credibility degree. The novelty of our approach concerns the modeling of the credibility of the user, through his/her trust and commitment in the social network. A social recommender system based on this approach is developed and experiments have been conducted using the Yelp social network. The evaluation results demonstrated that the proposed hybrid approach improves the accuracy of the recommendation compared with the user-based CF algorithm and solves the sparsity and cold start problems.     

SOMAR: A SOcial Mobile Activity Recommender

A 2010 survey (Nielsen) showed that 22.7% of the time spent on the Internet is on a social network. Moreover, there is an increasing demand to access social networks by mobile phones, i.e., around 30% globally. Social networking has become a reality, and it generates an incredible amount of information that is sometimes difficult for users to process, especially from mobile phones. Several links, activities, and recommendations are proposed by networked friends every hour, which together are nearly impossible to manage. There is a need to filter and make accessible such information to users, which is the motivation behind developing a mobile recommender that exploits social network information. Thus, in this paper, we propose the design and the implementation of a SOcial Mobile Activity Recommender (SOMAR) that can integrate Facebook social network mobile data and sensor data to propose activities to the user (e.g., concert or computer science seminar). The recommendations are completely calculated in situ in the mobile device with an embedded data mining component. We analyze how to compute and update the social graph in case of changes in social relationships or user context. The paper also presents a case study to analyze the performance of the method.     

Data dissemination using interest-tree in socially aware networking

Socially aware networking (SAN) exploits social characteristics of mobile users to streamline data dissemination protocols in opportunistic environments. Existing protocols in this area utilized various social features such as user interests, social similarity, and community structure to improve the performance of data dissemination. However, the interrelationship between user interests and its impact on the efficiency of data dissemination has not been explored sufficiently. In this paper, we analyze various kinds of relationships between user interests and model them using a layer-based structure in order to form social communities in SAN paradigm. We propose Int-Tree, an Interest-Tree based scheme which uses the relationship between user interests to improve the performance of data dissemination. The core of Int-Tree is the interest-tree, a tree-based community structure that combines two social features, i.e., density of a community and social tie, to support data dissemination. The simulation results show that Int-Tree achieves higher delivery ratio, lower overhead, in comparison to two benchmark protocols, PROPHET and Epidemic routing. In addition, Int-Tree can perform with 1.36 hop counts in average, and tolerable latency in terms of buffer size, time to live (TTL) and simulation duration. Finally, Int-Tree keeps stable performance with various parameters.