feat: Add TikTok/Douyin platform support (video feed, livestream, e-commerce)

[akz142857]

Summary

I'd like to propose adding TikTok/Douyin as the third simulation platform in OASIS, alongside Twitter and Reddit. This would make OASIS the first open-source framework capable of simulating TikTok-style algorithmic-feed social commerce at scale.

Motivation

• TikTok/Douyin is the world's largest short-video platform (7-8B DAU), but no open-source multi-agent simulation framework exists for its ecosystem
• OASIS's architecture (ActionType enum + Platform getattr dispatch + pluggable recsys) is well-suited for platform extension
• Demand exists: the OASIS paper mentions scalability across platforms, and multimodal support is an open request ([Feature Request] Multimodal Support to Agent Module #41, [Feature Request]Plan to support multi-modal #47)
• Use cases: content virality prediction, livestream commerce simulation, recommendation algorithm research

Proposed Changes

1. ActionType Extension (typing.py)

14 new TikTok-specific actions:

• Feed: UPLOAD_VIDEO, WATCH_VIDEO, SHARE_VIDEO, DUET, STITCH, NOT_INTERESTED
• Livestream: ENTER_LIVESTREAM, EXIT_LIVESTREAM, LIVESTREAM_COMMENT, SEND_GIFT, START_LIVESTREAM, END_LIVESTREAM
• E-commerce: VIEW_PRODUCT, ADD_TO_CART
• Plus get_default_tiktok_actions() reusing 5 existing actions (LIKE_POST, CREATE_COMMENT, FOLLOW, PURCHASE_PRODUCT, SEARCH_POSTS, TREND, DO_NOTHING)

2. Traffic Pool Racing Recommendation (recsys.py)

TikTok's core algorithmic differentiator:

• 7-level traffic pool with percentile-based promotion/demotion (top 20% promoted, bottom 30% stopped)
• Scoring: completion_rate (0.35) + like_rate (0.15) + comment_rate (0.15) + share_rate (0.20) - negative_rate (0.15)
• Feed composition: 70% interest-matched + 15% following + 10% explore
• 72-hour exponential time decay
• Per-user interest personalization via topic tag matching
• All 14 algorithm parameters configurable via kwargs

3. Platform Handlers (platform.py)

14 async methods following the existing (self, agent_id, message) dispatch pattern, with trace recording and DB operations.

4. Conditional DB Schema (schema/)

4 new tables created only when platform == "tiktok":

• video (traffic pool metadata, completion rates)
• livestream (room state, viewer counts, GMV)
• livestream_viewer (session tracking with stay duration)
• orders (e-commerce conversion tracking)

5. Agent Graph Generator (agents_generator.py)

generate_tiktok_agent_graph() — loads JSON profiles with TikTok-specific fields (role, interested_topics, follower_count, video_count). Appends structured interest tags to bio for recsys personalization.

6. Environment (env.py)

DefaultPlatformType.TIKTOK branch with tiktok_recsys_params passthrough for configurable algorithm tuning.

Implementation Status

I have a working implementation on a fork: akz142857/oasis@feat/tiktok-platform

• 13 files changed (9 modified + 4 new SQL schemas)
• ~1,200 lines added, fully additive (no changes to existing Twitter/Reddit behavior)
• All platform handlers follow existing code conventions

Still needed before PR submission:

• Unit tests for TikTok actions (following test_user_create_post.py pattern)
• Update test_action_docstring.py and test_twitter_user_agent_all_actions.py
• Example script in examples/
• Documentation in docs/key_modules/actions.mdx

Questions

1. Does this direction align with OASIS's roadmap? Is the team open to a third platform?
2. Should TikTok actions be added directly to ActionType or would a separate action enum be preferred?
3. Any preferences on the recsys approach (new function vs. subclassing)?

Happy to adjust the design based on feedback before submitting the PR.

[lightaime]

hi @akz142857. That would be amazing. maybe @echo-yiyiyi can help with coordinating this

[akz142857]

Sounds good, I’ll start working on it and submit a PR for review once it’s ready. @lightaime

[echo-yiyiyi]

@akz142857 Thanks very much for your idea. Let me know if you need any further help.

[akz142857]

PR #217 is up and linked to this issue, and I’ve addressed the current review feedback. Happy to iterate further if needed.
@echo-yiyiyi