Moonton Games

Scenario

As a MOBA game heavily reliant on teamwork, MLBB faces extremely complex text security challenges:
1. Hyper-Complex Linguistic Environment: Operating in 130+ countries and 11+ languages (e.g., Indonesian, Filipino, Burmese, Turkish), the platform struggles with abbreviations and slang unique to minor languages.
2. Toxicity in Competitive Play: High-intensity competition often triggers verbal abuse, regional discrimination, and provocation, leading to player churn and a toxic gaming environment.
3. Threats to Commercial Interests: Public channels are frequently targeted by illegal "top-up" ads, fraud, and RMT (Real Money Trading) redirections, threatening the game’s economic ecosystem and player security.

Solutions

DeepCleer developed a multi-dimensional real-time text risk control solution for MLBB:
1. Native Multilingual Moderation Models: Instead of simple translation, native models for minor languages (Indonesian, Thai, Arabic, etc.) were deployed to accurately identify localized variants, abbreviations, and homophones.
2. Deep MOBA Context Optimization: Established an exclusive MOBA vocabulary to recognize game-specific slang and insults, such as "beatrix nup" (noob) or "tank kau ntol" (insulting slang), which are often missed by generic filters.
3. Scenario-Based Tiered Management:- In-Game Chat: Millisecond-level interception of verbal abuse and regional discrimination to purify the competitive environment.- World Channel: High-precision identification of hidden contact info and prohibited advertisements to prevent traffic leakage and fraud.
4. Granular Labeling System: Categorized abusive behavior into sub-labels such as personality insults (intellectual/filth/animal-based), sexual profanity, and behavioral slangs, enabling customized enforcement.