Analysis of social unrest events using social network forensics

Abstract

In the past few years, there have been many social unrest events in many countries. People mainly use social networks to spread events and exchange views on them. Social network forensics focuses on collecting and analyzing the content and user information of criminal cases with evidence on social media. However, it is difficult to manually analyze large amounts of open-source information. These data must be processed and filtered automatically in advance. In this thesis, we propose a social unrest event detection model using social network forensics. Specifically, we analyze the social unrest events on the following three processes: event-related topic extraction, event detection, and user analysis. In the experimental part, we mainly utilize the posts on the Lihkg discussion forum, which is a prominent forum in Hong Kong, created from 1 Aug 2019 to 31 Aug 31 2020. First, the event-related topic model aims to automatically extract keywords and topics from massive social media data. Considering social media data, we focus on automatically extracting the temporal and spatial topics of social unrest events from a large amount of open-source data. A temporal and spatial topic model is proposed to vectorize the characters using character embedding, then vectorize the topics using word segmentation methods, and finally construct the topic model based on the MLP (Multi-Layer Perceptron) algorithm. Second, the event detection process is to detect information related to social unrest events. An entity-based integration event detection model is proposed for event extraction and analysis in social media. The framework integrates two modules. The first module includes named entity recognition technology utilizing BERT (Bidirectional Encoder Representation from Transformers) algorithm to extract the event-related entities and topics of social unrest events during social media communication. The second process suggests the K-means clustering method and DTM (Dynamic Topic Model) for dynamic analysis of these entities and topics. In addition, the comparative experiment is performed to reveal the differences between Chinese users on Lihkg and Twitter for comparative social media studies. Finally, we analyze the users who have posted event information on the social network. The user analysis process includes user analysis and community detection. Community detection is a clustering method for graph and networks to identify communities and reveal aggregating behaviors in the network. It has become a crucial task in social network forensics. Our proposed user analysis and community detection model utilizes the GCN (Graph Convolution Network) algorithm and can analyze the social unrest events considering user attributes and communication content. The results prove that our social unrest event detection model can detect social unrest events and correctly extract keywords and entities in social networks that somehow indicate the characteristics of social unrest events. The model can also offer some essential insights into considering both user behavioral attributes and content features of communication to detect user community and community structure. In the analysis of social unrest events using social network forensics, we find that the information filtered from social networks can be utilized to analyze social unrest events, accurately identify key members on social networks, and even develop corresponding tactics.