Recommendation Algorithm Feeder

English | 中文

A research-oriented tool for simulating user behavior to study and influence video recommendation systems.

An open-source framework to explore the behavior of video platform recommendation algorithms by simulating structured user interactions. Designed for personal feed customization, testing, and algorithm transparency research.

Features

• Multi-platform Support: Bilibili, YouTube (expanding)
• Smart Matching: Keyword-based video filtering with multiple matching modes
• API Mode: Direct platform API calls for efficiency and stability
• Simulated Playback: No actual video playback required, configurable speed and duration
• High-Performance Concurrent Processing: Multi-threaded processing with configurable concurrency
  • Performance Example: 10 concurrent + simulated playback = 121.5 minutes of video processing in just 2 minutes
• Active Search: Automatic keyword search when recommendation effectiveness is poor
• Smart Queue: Automatic video processing queue management, avoiding duplicates
• Multi-source Recommendations: Support for homepage, related, shorts and other content sources
• History Tracking: Complete viewing history and matching statistics
• Flexible Configuration: 50+ configurable parameters for different use cases

Project Structure

recommendation-algorithm-feeder/
├── src/
│   ├── core/                    # Core architecture
│   │   ├── feeder.ts           # Main controller
│   │   └── logger.ts           # Logging system
│   ├── platforms/              # Platform adapters
│   │   ├── bilibili/           # Bilibili implementation
│   │   ├── youtube/            # YouTube implementation
│   │   └── platform-factory.ts # Platform factory
│   ├── services/               # Business services
│   │   ├── concurrent-player.ts        # Concurrent processing
│   │   ├── simulated-playback-manager.ts # Simulated playback
│   │   ├── history-service.ts           # History tracking
│   │   └── match-service.ts             # Matching service
│   ├── matchers/               # Matching algorithms
│   ├── config/                 # Configuration management
│   ├── utils/                  # Utility functions
│   └── types/                  # Type definitions
├── tests/                      # Test code
├── docs/                       # Documentation
├── cookies/                    # Cookie storage
├── history/                    # Historical data
└── logs/                       # Log files

Quick Start

1. Install Dependencies

npm install

2. Configure Environment

cp .env.example .env
# Edit .env file with your target keywords and settings

# Create cookies directory and add your platform cookies
mkdir cookies
# Place your Bilibili cookie in: cookies/bilibili.txt
# Place your YouTube cookie in: cookies/youtube.txt

3. Run Program

npm run dev

For detailed usage instructions, see docs/usage.md

Core Configuration Options

Basic Configuration

• Platform Selection: PLATFORM - Supports bilibili/youtube
• Target Keywords: TARGET_KEYWORDS - Comma-separated keyword list
• Match Mode: MATCH_MODE - any (match any keyword) or all (match all keywords)
• Execution Rounds: BROWSE_COUNT - Total execution rounds
• Play Duration: PLAY_DURATION - Video playback duration per video (seconds)

Efficiency Mode

• API Mode: USE_API_MODE - Direct platform API calls, no browser rendering needed
• Simulated Playback: USE_SIMULATED_PLAYBACK - Simulate playback behavior, save bandwidth
• Playback Speed: SIMULATED_PLAYBACK_SPEED - Simulated playback speed multiplier (1-5x)
• Concurrent Count: CONCURRENT_PLAYERS - Number of videos processed simultaneously (1-10+)
• Ultra-High Performance: With simulated playback, process 60x more content in the same time

Smart Features

• Active Search: ENABLE_ACTIVE_SEARCH - Automatic search when recommendations are ineffective
• Search Threshold: ACTIVE_SEARCH_THRESHOLD - Consecutive unmatched rounds to trigger search
• Queue Management: MAX_VIDEOS_PER_QUEUE - Maximum videos processed per round
• Video Source Switching: INITIAL_VIDEO_SOURCE - home/related/short

System Configuration

• Headless Mode: HEADLESS - Run in background without browser interface
• Log Level: LOG_LEVEL - error/warn/info/debug
• History Retention: HISTORY_CLEANUP_DAYS - Days to retain historical data
• Proxy Support: HTTP_PROXY/HTTPS_PROXY - Network proxy configuration

Architecture Design

The project uses modular design for easy extension:

• Core Module: AlgorithmFeeder main controller, coordinates all components
• Platform Interface: PlatformInterface abstraction layer, supports Bilibili/YouTube extension
• Matching Service: MatchService intelligent keyword matching and filtering
• Concurrent Processing: ConcurrentPlayer multi-threaded video processing
• Simulated Playback: SimulatedPlaybackManager efficient playback simulation
• History Service: HistoryService complete data recording and statistics
• Configuration Management: Unified configuration loading, validation and wizard
• Logging System: Hierarchical log output and file recording

Usage Instructions

For detailed configuration and usage methods, please refer to:

• Complete Usage Guide - Detailed configuration instructions

Disclaimer

This tool is for personal learning and research purposes only, intended for actively adjusting personal recommendation algorithm preferences. Please comply with relevant platform terms of service and use responsibly.

License

AGPL-3.0 License