社会网络分析与挖掘若干关键技术研究

With the rapid development of the Internet and Mobile Network, social network are changing people's traditional social life and information-access patterns. However, the increasing popularity of social network brings along an enormous amount of users and user generated contents, as well as complex social relationships, which also brings new opportunities and challenges for industry and academia. To tackle the key issues in social network analysis and mining, we will develop methods and models based on data mining and machine learning techniques. The main tasks of this project can be concluded as follows. (1) Summarizing the social hot spot and monitoring online community critical information based on topic mining. (2) Modeling user preference based on the user historical behaviors (3) Analyzing social relationship network and predicting social links. (4) Studying learning to rank algorithms to optimize the information retrieval and recommendation in social network. Thus, our work has practical significance in various application scenarios, such as tracking and monitoring public opinions, personalized information services, and Customer Analysis as well as Precision Marketing.

随着互联网、移动网的发展与普及，社会网络正在颠覆人们传统的社交生活以及信息索取模式。社会网络繁荣的同时也导致了庞大的用户规模、海量的内容信息以及复杂的社交关系。这为工业界与学术界带来了新的机遇与挑战。本项目将针对目前社会网络分析与挖掘中的若干关键技术问题，开展基于数据挖掘与机器学习的相关理论方法和模型研究。具体包括：(1)研究社会网络上的主题挖掘技术，提炼社会热点，把握社区动态；(2)研究用户历史行为，挖掘用户兴趣并建模行为方式；(3)研究社交关系网络结构，分析并预测社会关系；(4)研究基于排序学习的搜索查询，优化社交推荐。本课题的成果可望应用于社会舆论跟踪与监控、个性化信息服务、客户分析与精准广告等众多方面，具有广阔的应用前景。

项目在社会网络主题挖掘、社会网络用户行为建模与社交推荐、社会网络关系建模与预测、排序学习模型等方面进行了深入研究，提出了排序学习中的特征选择、基于多层卷积神经网络的特征学习方法，基于轨迹大数据的目的地预测、最优位置查询算法，基于知识图谱的语义分析模型、优化评价指标的多任务学习框架，社交网络下的情感分类、隐式情感检测方法等，并得出了若干有价值的结论。.项目在国际国内重要期刊和本领域重要的国际会议上发表论文 27篇，包括ACM TODS、IEEE TNNLS、TIP、TC、TAE、TITS及国际著名会议ACL、NIPS、CVPR等，其中，CCF Rank A 类期刊和会议6篇，CCF Rank B类期刊和会议10篇，国内一级学报（如软件学报等）论文3篇；培养研究生12名；受理发明专利10项；获得学术竞赛奖2项。