Abstract
Social network analysis is becoming more relevant with the rise of Large Language Models and other AI frameworks, particularly as online content creation surges. However, analyzing social network data from different platforms to study online opinions and influence campaigns remains challenging. Even more complex is the concept of multi-source social network analysis. Knowledge graphs, an essential part of Web 4.0 and crucial in information modeling and encoding, offer a unique solution. They allow for a more concise encoding of relationships between entities and related discussions. However, there is limited investigation into using multi-source data for examining online information campaigns and identifying influential opinions, entities, and topics. This research introduces a multisource knowledge graph model, KG-CFSA (Knowledge Graph Contextual Focal Structure Analysis), built using a Cartesian merge. The Cartesian merge establishes relationships across multiple documents, entities, and topics using a Cartesian product, incorporating WikiData and DiffBot data. This method creates multi-layer and multi-relationship heterogeneous graphs. To enhance the identification of contextual focal structures in multi-source social network data with dynamic relationships, Laplacian and non-Laplacian matrices are used. These can be used to study information across multi-source social network data from platforms such as Reddit, X (formerly Twitter), YouTube, and blogs. Using a Laplacian matrix and measuring graph similarity via their diagonal matrix enables the exploitation of near and distant relationships, effectively establishing similarity in a multi-source knowledge graph. To validate the hypotheses with real-world data, a multi-source knowledge graph is created using a dataset from the Indo-Pacific region, focusing specifically on the Belt and Road Initiative. The multi-source knowledge graph was then subjected to KG-CFSA and knowledge embedding. The KG-CFSA identified key information, sources, and entities driving the conversation. Knowledge embedding was accomplished by applying knowledge graph embedding and using four scoring mechanisms: ComplEx, TransE, DistMult, and HolE, on a domain knowledge graph of the Indo-Pacific Belt and Road Initiatives.Visit Publisher
