| Session Name: | Machine Learning Summit: Aegis Engine: Building Multi-modal Moderation System in NetEase Games |
| Speaker(s): | Yunbo Peng |
| Company Name(s): | NetEase Games AI Lab |
| Track / Format: | Machine Learning Summit |
Did you know free users get access to 30% of content from the last 2 years?
Get your team full access to the most up to date GDC content
| Overview: | User-generated content (UGC) and social interaction greatly improve player participation and game entertainment. To keep the games free from inappropriate content, such as eroticism, violence, spamming, abuse, etc., building a content moderation system is of vital importance. However, manual moderation is time-consuming, laborious and costly.nThis session introduces the Aegis Engine, a novel multi-modal content moderation system which applies deep learning techniques for not only text but also unstructured image and audio data. The engine contains three sub-systems. Image sub-system exploits fine-grained recognition and real-time OCR algorithms. Audio sub-system is built upon keyword-enhanced and noise-robust speech recognition. Text sub-system contains word embedding based inappropriate content mining method. The Aegis Engine processes more than 7,000,000 images and 100,000 hours of audio data every day, covering almost all the games in NetEase Games such as Knives Out, Onmyoji, and LifeAfter. |
Recommended
