Modelling news dissemination networks using a community-based graph traversal algorithm and its performance evaluation

In this study, a directed weighted graph containing 500,000 user nodes and 100,000 news records was constructed.Node attributes are labelled with user activity and news sentiment tendency.Edge weights were determined according to propagation time attenuation (5%/hour) and interaction frequency.The community was divided by Louvain's algorithm, and the core nodes were identified by fusing node betweenness centrality and PageRank.Path traversal was optimised with 0.2 restart probability.Compared with traditional methods such as the shortest path algorithm and static community random walk, CGTA achieves 92.3% (76.6% for traditional methods) and 89.1% for core node recall (71.4% for traditional methods), which are 15.7% and 17.7% higher respectively.The structural equation model quantified that user activity (38.2%), emotional tendency (29.5%), and propagation time (22.3%)dominated the path formation, and the propagation speed decreased by 12.3% (r = -0.78)for every 1 hop increase in path length, and the core node expansion effect reached 67.5%.The study integrates multi-source data from Weibo and Toutiao, and cross-validates it with 5% to confirm the effectiveness of the algorithm in breaking/regular news and communities of different sizes.