Use the GDC 2021 Session Viewer to browse and sort the ever-growing list of sessions by pass type, topic, and format. You will be able to build your personal schedule directly within the event platform, once live early July 2021.
Wuhe Zou (AI Engineer, Netease Game)
Pass Type: All Access Pass, Core Pass
Topic: Programming
Format: Session
Vault Recording: TBD
Audience Level: N/A: I would like to deliver a virtual presentation only
This presentation consists of two parts, Game Environment Guardian and Game Public Opinion Analysis System. Game Environment Guardian includes text detection algorithm, voice detection algorithm and illegal expression mining algorithm. The user expression can be quite different between games. And the user expression change as the system evolves. The presentation introduces how they solve above problem with the algorithms and system, quantitative evaluation results, and application situation.
Game Public Opinion Analysis System includes web crawler module, data flow module, emotion analysis module, hotspot mining module. This presentation will introduce how we set up the whole system framework.
Overall, the presentation proposed a Game Environment Guardian System, which cover text chat, voice chat and player behavior. Especially the proposed voice chat detection method is 6~10 times quicker than traditional ASR method. The text detection algorithm has high precision (95%+), high speed (5000+qps) and cover 10+ illegal categories. Furthermore, the active learning architecture make the algorithm constant iterate with players' expression changing. The AI x Game Environment Guardian System has been used in 8 different games.
This presentation also proposed a Game Public Opinion Analysis System, which covers major player community. The system has a variety of functions, such as hotspot mining, emotion analysis, automatic opinion aggregation, public opinion alarm push. And the system has been used by over 20 games.
This presentation shares our experience with applying AI algorithm to guard game environment, including a Highly time-sensitive voice detection algorithm, a precise text detection algorithm and a active learning architecture (illegal expression mining algorithm). Also this presentation share our experience with setting up a public opinion analysis system with hotspot mining, emotion analysis and automatic opinion aggregation.
The target audiences are AI Engineer/game programing developer or anyone concerned about game environment guardian (abuse, eroticism, illegal advertisement, screen flooding, real-money-trade). Some basic concept of artificial intelligence algorithm will be better.