XP-STM: A cross-platform sentiment transferability model for negative public sentiment identification and mitigation

초록

Social networks operate as interconnected digital ecosystems where negative sentiments propagate rapidly across platforms, transforming in intensity as they migrate between communities. Conventional sentiment analysis methods examine platforms in isolation, failing to capture cross-platform sentiment evolution and cascade dynamics. This study introduces the Cross-Platform Sentiment Transferability Model (XP-STM) to systematically analyze negative sentiment migration pathways, identify platform-dependent transformation patterns, and determine structural factors that amplify or suppress negativity during cross-platform movement. The framework employs three analytical layers: cross-platform integration using multilingual transformer-based architecture for unified sentiment representation, influencer node identification through node2vec embeddings combined with network centrality metrics, and cascade dynamics modeling using XGBoost ensemble methods for transferability prediction. The model extends Node Expression Capacity calculations to accommodate platform-specific constraints and temporal synchronization across heterogeneous environments. Results are achieved through computational implementation using distributed training on 8 & times; NVIDIA A100 GPUs across 2.3 million posts from Twitter, Reddit, Weibo, ResearchGate, Zhihu, and Sentiment140 datasets. Performance evaluation demonstrates XP-STM achieves F1-scores ranging from 0.87 to 0.92 across all platforms, cross-platform transferability prediction AUC of 0.89-0.95, Counter-Messaging Efficacy of 40-65 %, and Cascade Containment Index exceeding 0.60, consistently outperforming baseline models EANN, AOAN, and HMCan.

키워드