Similar users

A similar user query returns users who are most similar to a given user. Common applications include finding users with similar preferences, building user segments, and enabling social features.

You can do user similarity in two ways:

1. Collaborative similarity, which finds users with similar interaction patterns - which users interact with the same items?
2. Content similarity, which finds users with similar profile attributes (e.g., interests, demographics).

Collaborative similarity

Prerequisites

1. An engine with a trained collaborative embedding (e.g., ALS) configured for users
2. A user to find similar users for

Query example

To find similar users based on collaborative filtering (interaction patterns), use the similarity retrieve type with a precomputed user embedding:

{
  "query": {
    "type": "rank",
    "from": "user",
    "retrieve": [
      {
        "type": "similarity",
        "embedding_ref": "als_user_embedding",
        "query_encoder": {
          "type": "precomputed_user",
          "input_user_id": "$parameters.user_id"
        },
        "limit": 20
      }
    ]
  },
  "parameters": {
    "user_id": "user123"
  }
}

Content similarity

Prerequisites

1. An engine with a content-based user embedding (e.g., text embedding from user profiles) configured
2. A user to find similar users for

Query example

To find similar users based on profile attributes (e.g., interests, bio text), use the similarity retrieve type with attribute pooling:

{
  "query": {
    "type": "rank",
    "from": "user",
    "retrieve": [
      {
        "type": "similarity",
        "embedding_ref": "text_user_embedding",
        "query_encoder": {
          "type": "user_attribute_pooling",
          "input_user_id": "$parameters.user_id"
        },
        "limit": 20
      }
    ]
  },
  "parameters": {
    "user_id": "user123"
  }
}