A Knowledge Graph-Based Early Warning Model for Network Nuisance Actors

With the rapid development of social media and networking platforms, nuisance information disseminated online has a significant negative impact on daily life. This data is characterised by multimodality, fragmentation and a lack of labelling, making it challenging to aggregate and analyse raw data, identify nuisance information and actors, and discover potential nuisance events. To address this issue, this paper presents a knowledge graph-based early warning model for identifying network nuisance subjects. Firstly, we construct a multimodal knowledge graph for network nuisance governance, based on the characteristics of network nuisance data. This allows us to align multi-source data and integrate redundant information through structured data mapping and attribute fusion. Secondly, we establish a nuisance labelling system from three dimensions: user attributes, behavioural habits, and interest areas. We then combine the K-Means algorithm to construct a group profile of network nuisance. Thirdly, to address the issue of potential subjects being strongly concealed, we design a social influence assessment model that integrates improved PageRank and KShell algorithms. Combined with the similarity of portraits, we propose an early warning method based on threat level to accurately identify potential nuisance subjects. Experiments show that the model achieves an accuracy rate of 83.3 % in predicting potential network nuisance subjects, demonstrating its practical application value.