Overview
Generate personalized Polymarket market recommendations for a wallet address. This endpoint uses user behaviour modeling to deliver tailored prediction market recommendations through a three-step process: user profiling, candidate generation, and personalized reranking.
Key Concepts
- User Behaviour Modelling: Uses onchain betting history from Polygon to profile user behaviour and cluster them. This is done in real-time so that the user features and embeddings are up to date.
- Candidate Generation: When a request to the API is received, user features are fetched from a hot-user vector and made available to come up with relevant markets. This step can be controlled by passing filters (like start date, close date, daily volume, daily price changes, etc).
- Reranking: Uses a Large Recommendation Model (LRM) to personalize and reorder the candidates generated in step 2. The current model being trained and provided in this alpha version for testing should be defined in filters.scoring field of the request body. See the RecommendationFilters schema for detailed algorithm descriptions and options.
- Market Filtering: Apply sophisticated filters including volume metrics, liquidity thresholds, market status, tags, and date ranges to control candidate generation
- Ordering: Sort candidates by volume metrics, liquidity, time-based fields, price movements, or market quality indicators
- Metadata: Optionally include full market details including prices, volumes, dates, and recommendation reasons
Usage
Returns a list of recommended Polymarket markets for a specific wallet address - markets the user is most likely to be interested in trading. The system first profiles the user based on their onchain betting history, then generates relevant market candidates using the provided filters, and finally reranks them using the specified scoring algorithm. Use the filters parameter to customize ordering, apply volume/liquidity thresholds, filter by tags or dates, and choose between raw Polymarket API ordering or ML-powered personalized reranking. The return_metadata parameter controls whether full market details are included in the response.