Skip to main content

Recommendation Architecture

This document outlines the architecture for the SplitScreen user recommendation engine. The system is designed to be flexible, catering to two primary client types:

  1. Single-Game Clients: Typically a game studio that wants to provide social matchmaking for their specific game.
  2. Multi-Game Clients: Platforms like gamer hubs (e.g., Shonen) where users have profiles with multiple games.

The recommendation engine is built around the GenericUserProfile model.

Core Recommendation Factors​

These are the primary factors used to calculate a match score between two users, based on the current GenericUserProfile model.

1. Games & In-Game Progress​

  • Logic: The primary factor for matching. The scoring logic adapts based on the client type.
  • Adaptation:
    • Multi-Game Clients: The score increases with the number of games two users have in common.
    • Single-Game Clients: This becomes a simple check for the one shared game. The recommendation score will then heavily rely on the in-game metrics below.
  • Metrics:
    • PlaytimeMinutes: A strong indicator of experience. The system will reward users with similar playtime.
    • AchievementsCount: A proxy for skill and dedication. Similar counts suggest a better match.

2. Genres​

  • Logic: Shared taste in genres is a good indicator of compatibility.
  • Adaptation:
    • Multi-Game Clients: This is a powerful signal. Overlap in preferred genres will significantly boost the match score.
    • Single-Game Clients: This factor has less weight, as all users are technically playing the same genre. However, it can still be useful if the user has provided a broader list of genre preferences.

3. Availability & Schedule​

  • Logic: Critical for social matchmaking. Users must be able to play at the same time.
  • Metrics:
    • Schedule.Timezone: Used for a baseline grouping of users.
    • Schedule.Slots: The system will perform an overlap analysis on users' available time slots to calculate a "schedule compatibility score."

4. Location & Language​

  • Logic: Determines network latency and the ability to communicate.
  • Metrics:
    • LocationInfo.CountryCode / Region: Used to estimate latency. Users in the same region receive a higher score.
    • Note: CountryCode can be a proxy for language, but an explicit language field is recommended for better accuracy (see enhancements below).

5. Personality (Optional)​

  • Logic: Gauges social compatibility. This is an optional but powerful factor. The documentation should emphasize that providing this data leads to better recommendations.
  • Metrics:
    • MBTI: Uses the Myers-Briggs Type Indicator. A scoring matrix will be used to compare two MBTI types.
    • Yee's Gamer Motivation Model: This model can also be supported to provide a different lens on player personality and motivation.
  • Note: If a user has not provided this information, it will not negatively or positively impact their score.

Scoring & Ranking​

The recommendation engine will use a weighted scoring system. Each of the factors above will be assigned a weight, which can be adjusted based on client feedback and performance analysis. Clients can configure these weights for themselves using the API.

  • Initial Weights (Example):
    • Games: 30%
    • Skill/Rank: 25%
    • Availability: 20%
    • Playstyle: 10%
    • Location/Language: 10%
    • Personality: 5%

The final output will be a ranked list of users, ordered by their total match score.

Data Flow​

  1. User Data Ingestion: Clients send user profiles via the /v1/users endpoint.
  2. Asynchronous Processing: The system asynchronously processes new and updated user profiles.
  3. Pre-computation: To ensure low-latency recommendations, the system will pre-compute potential matches.
  4. Recommendation Request: When a client requests recommendations via /v1/users/{user_id}/recommendations, the system retrieves the pre-computed, ranked list.
  5. Real-time Filtering (Optional): The system can apply real-time filters to the pre-computed list (e.g., "recommend me users only for a specific timezone or users who play a specific game.").