Presentation Information
[5Yin-A-06]Dynamic User Profiling in GaaS Community Backlash: A Hybrid Framework for Risk Detection and SegmentationDifferentiating Core Players from External Interference in Genshin Impact
〇Yanming He1,2, Hideaki Takeda2,1 (1. Graduate University for Advanced Studies, SOKENDAI, 2. National Institute of Informatics)
Keywords:
Games-as-a-Service,Computational Social Science,User Profiling,Behavioural Log Analysis,Risk Detection
In the Games-as-a-Service (GaaS) model, community health is critical for user retention. However, backlash events often obscure genuine feedback, blurring the distinction between constructive complaints from core players and disruptive noise from tourists. This study introduces a training-free dynamic user profiling framework to automate risk detection and segmentation during such crises.
Leveraging a massive dataset from Genshin Impact, we propose a hybrid methodology integrating behavioural log analysis with semantic text mining. Specifically, we utilise Differentially Weighted Reception Importance (DWRI) and HDBSCAN clustering to map user patterns. This approach successfully identifies distinct cohorts and quantifies divergent discourse patterns—where external narratives destabilise community norms.
Results demonstrate the framework effectively decouples high-risk opportunists from core segments without relying on labeled data. This research offers a scalable tool for churn prediction, community governance, and risk management, bridging computational social science with practical game operations.
Leveraging a massive dataset from Genshin Impact, we propose a hybrid methodology integrating behavioural log analysis with semantic text mining. Specifically, we utilise Differentially Weighted Reception Importance (DWRI) and HDBSCAN clustering to map user patterns. This approach successfully identifies distinct cohorts and quantifies divergent discourse patterns—where external narratives destabilise community norms.
Results demonstrate the framework effectively decouples high-risk opportunists from core segments without relying on labeled data. This research offers a scalable tool for churn prediction, community governance, and risk management, bridging computational social science with practical game operations.