社交网络中基于图挖掘的用户行为分析与异常检测技术研究

Currently rapid advance of social networks facilitates social information dissemination while are being bothered a variety of issues, e.g., anomalies in user accounts threat information security of normal users and hinder the healthy development of social networks. Network topology constructed by information on users behavior contains rich behavior information. It is a technical challenge, which is imperative to be solved, for valuable knowledge extraction, anomalous behavior detection as well as dynamically random allocation and adaptation for user nodes based on the practical situations. Based on such motivations, the project begins with feature extraction for user behavior, label propagation and graph pattern mining, and focus on anomaly detection and user behavior analysis with massive user behavior data in social networks, which are associated by noise points restoring strategy in data preprocessing and optimization schemes in large-scale graph mining. Hence, the applications of graph mining techniques in anomalous behavior detection can be effectively promoted, which is in line with the current trend in behavior mining and research on social networks.

社交网络的飞速发展为社会信息的传播带来便利的同时，也饱受着各种问题的困扰，例如用户账号异常等问题不仅严重威胁着正常用户的信息安全，也影响了整个社交网络的健康发展。用户行为信息构成的网络拓扑蕴涵着丰富的行为信息，如何高效准确得运用数据挖掘的方法来从这些信息中淬炼出有价值的知识，进而探测出异常的行为，根据实际情况对节点进行动态随机分配并相应的进行自适应调整也是在实际发展过程中亟需解决的技术难点。基于以上出发点，本项目拟从用户行为的特征提取、标签传播、异常事件挖掘等研究点手入手，结合数据预处理阶段的噪声点修复策略以及大规模图挖掘等优化方案，对社交网络中的海量用户行为数据进行异常检测及行为分析，从而有效促进图挖掘技术在异常用户行为领域的研究和应用，符合当下针对行为挖掘和社交网络研究的发展趋势。

用户行为信息蕴涵着丰富的语义信息，如何高效准确地运用深度学习和数据挖掘的方法来从这些信息中淬炼出有价值的知识，进而探测出异常的用户行为，从而在实际网络中加以调整和应用具有很好的研究价值和应用价值。本课题对社交网络中的海量用户行为数据进行了行为分析及异常检测，具体研究内容包括：结合用户结合链接和内容的社交网络社区的发现算法研究；基于深度学习的用户消费行为分析与预测研究；基于多模态语义提取和情感注意力加强机制的用户情感分析与研究；基于可变兴趣模型的用户兴趣挖掘研究；基于改进的自编码器的异常账号检测。通过上述研究内容，对社交网络中的视频、文本、图像等内容进行了过滤和审查，促进了社交网络的健康发展。