基于词语动能聚类的社会化媒体热点话题发现建模与优化方法

Modeling and Optimization of Hot Topic Discovery in Social Media Based on Clustering of Word Kinetic Energy

摘要: 在社会化媒体情境下，社会化媒体舆情已成为社会舆情的新视域，社会化大数据分析正显现出日益重要的社会价值和商业价值。在新兴的社会化大数据管理领域，热点话题发现是网络舆情分析和数据治理基础而重要的课题，人们一直在研究和探索先进和适用的热点主题挖掘的理论和方法。针对传统的聚类算法用于微博话题检测时，存在特征向量过于稀疏和维度过高等问题，导致聚类结果不准确。本文通过对在话题传播周期中词语的突发性特征的研究，提出了一种基于传播周期的词语动能聚类(Word Kinetic Energy Clustering, WKEC)模型和算法。该文本聚类模型基于话题生命周期特性，引入物理学中的动能概念，用词语在话题爆发期的最大增长速度来表征词语的动能，并加入到词语权重的计算中，对经典的TF-IDF模型进行了改造。基于Single-Pass的算法设计和新浪微博真实数据集的实验结果表明，WKEC模型可以增强文本特征，提高话题发现的准确率。另外，由于微博话题实时性强，为了得到更接近真实的微博热点话题列表，本文在话题热度计算中引入衰减系数，并以爆发期尾部时间点作为话题热度开始衰减的时刻，给出了一种更加符合实际的话题热度计算方法。

Abstract: In the context of social media, social media public opinion has become a new perspective of social public opinion. Social big data analytics is showing an increasingly important social and business value. In the emerging field of social big data management, the discovery of hot topics in microblog is the basic and important issues for government public opinion analysis and data management; people have been studying and exploring advanced and applicable theories and methods of hot topic mining. However, when the traditional clustering algorithm is used in the microblog topic detection, the eigenvectors are too sparse and over-dimensioned, resulting in inaccurate clustering results. Therefore, this article explores the sudden features of words in the cycle of topic communication, and proposes the Word Kinetic Energy Clustering (WKEC) model and algorithm. The text clustering model, based on the topic life cycle feature, introduces the concept of kinetic energy theorem in physics, and calculates the kinetic energy of words with the maximum growth rate in the explosion period, which will be added to the weight of the words, modifying the classic TF-IDF model. Based on the algorithm design of Single-Pass and the real dataset from Sina Microblogs, the experimental results show that WKEC model can enhance the text features and improve the accuracy of topic discovery. In addition, due to the strong real-time of microblogging topic, in order to get closer to the real microblogging hot topic list, this article introduces the attenuation coefficient into the calculation of topic heat, and takes the tail time point of the explosion period as the decay moment of topic heat, proposing a more realistic method for the calculation of topic heat.