Design and Implementation of a Rule Engine for Content Risk Control

Introduction: The existing risk control system at Snowball community faced scalability and maintainability issues due to hard‑coded Scala/Java rules, lack of unified logging, and poor offline analysis capabilities.

Goal: Build a new rule engine that separates logic from data, supports high concurrency, provides a visual DSL for strategy configuration, stores snapshot data, and enables rapid response to emerging threats.

System Architecture: The engine consists of a scheduling core, feature sets, strategy sets, actions, and a logging engine. Features (facts) feed data into the context; strategies (conditions) are expressed using MVEL2; actions execute upon condition match. The engine supports both serial and parallel execution modes for features and strategies.

Key Components: Scheduler (process, rule, strategy, and resource sync), feature‑strategy integration, model feature unification, and a virtual environment for safe testing. The scheduling engine orchestrates component execution with versioned flows, allowing A/B testing and rapid iteration.

Implementation Details: A visual strategy editor provides a DSL that business users can manipulate without developer involvement. Strategies are written in a DSL that compiles to MVEL2 expressions; the following code snippet shows example rule definitions.

/** 策略1.1 */
rule "1.1"
    if
       ${date}>"2021-07-02" && funA(funB(${text})) && ${type}==1
    then
       action("1.1 matched");
       action("1.2 matched");
end

/** 策略1.2 */
rule "1.2"
    if
        ${date}>"2021-07-01" && ${type}==2 && ${list}.contain("rule")
    then
        action("1.3 matched");
end

/** 策略2.1 */
rule "2.1"
    if
        ${date}>"2021-07-03" && ${type}==4
    then
        action("2.1 matched");
end

Practical Deployment: Migrated 35 original rules, optimized feature reuse and parallel execution, reducing API calls from 34 to 14 and latency from 750 ms to 500 ms. The platform now supports rapid strategy adjustments, model integration, and comprehensive monitoring.

Conclusion: The rule engine improves risk detection efficiency, lowers development overhead, and provides a flexible, secure framework for future extensions in content moderation and other decision‑making scenarios.