社会网络的主题演化分析与传播趋势预测研究

Topic evolution analysis in online social networks is of great significance in theoretical research and application. However, the massive and low-quality data sampling, processing and computing are the key bottlenecks in it. Compressed Sensing (CS) theory is a new effective method for signal transmission and image compression, but it is not feasible for social networks directly. In this proposal, we first propose a theory which maps the tensor space of social network into time-frequency space of compressed sensing. By topic feature indicators scoring, sparse matrix designing and sparse-coding apriori, it will lay a theoretical foundation for applying compressed sensing theory into topic analysis. Then we present compressed sensing based strategies for high-quality topic extraction and evolution. For high-quality topic extraction, we design a local prior knowledge based reconstruction algorithm K-OMP, which can reconstruct the topic information of complete-approximation in the condition that the sampling data is far less than the original data set. For topic evolution, we design a dynamic redundant dictionary based evolution analysis algorithm DRD-EA to maintain the survival state of the topic stream. Finally, we develop the scale-free BA model to predict the topic communication trend based on the topic evolution state, topic communication intensity and node influence. The innovation results will be a new breakthrough to traditional topic extraction and analysis in large-scale social networks. It will help to mine the potential values in social networks more efficiently, therefore promoting its healthy development.

在线社会网络的主题演化分析有着重要的研究意义和应用价值，而其中主要的技术瓶颈是对海量低质数据的抽取与分析。基于压缩感知的数据采样与压缩理论，在信号处理领域已获得了广泛应用，但并不直接适用于社会网络。本项目中，首先通过特征指标打分、特征稀疏转化等，实现社会网络特征张量空间到压缩感知时频空间的映射，为压缩感知理论在主题分析中的运用提供依据。在此基础上，设计整合先验知识的主题信息重构算法K-OMP，在样本数据远少于原始样本集的情况下，重构逼近完整的主题信息，实现高质量主题抽取。进而，提出基于动态冗余字典的主题演化分析算法DRD-EA，分析主题流的内容和生命周期演化规律。最后，基于主题演化状态、传播强度和节点影响力等，改进无标度网络BA模型，对主题传播趋势进行预测。研究成果将是对传统的社会网络主题抽取与演化分析方法的全新突破，有利于更高效地挖掘社会网络数据的潜在价值，促进社会网络的健康发展。

本课题以在线社会网络的主题演化分析为背景，研究在线社会网络中海量低质消息的主题演化，并预测主题在社会网络中的传播趋势。在本课题中，首先通过特征指标打分、特征稀疏转化等，实现社会网络特征张量空间到压缩感知时频空间的映射。在此基础上，利用整合先验知识的主题信息重构算法K-OMP，在样本数据远少于原始样本集的情况下，重构逼近完整的主题信息，实现高质量主题抽取。进而，使用基于动态冗余字典的主题演化分析算法DRD-EA，分析主题流的内容和生命周期演化规律。最后，基于主题演化状态、传播强度和节点影响力等，对主题传播趋势进行预测。在开展高质量主题抽取与演化分析、主题传播趋势预测等方面的研究中，本课题创新性的提出了（1）社会网络张量空间向压缩感知时频空间映射的理论方法，实现更加准确高效地进行主题抽取与演化分析；（2）基于压缩感知的社会网络主题演化分析方法，实现高质量的主题抽取；（3）面向主题传播趋势预测设计主题传播演化模型BA-TEP，从而动态预测社会网络主题的传播趋势规律。