agentlans/FarisHijazi-kajiwoto.ai-chat

FarisHijazi/kajiwoto.ai-chat

Dataset Description

This dataset is a processed version of the FarisHijazi/kajiwoto.ai-chat dataset, containing extracted and re-formatted conversation data from Kajiwoto. Kajiwoto is a platform where users can create and interact with AI character datasets. This version focuses on providing the core conversational data in a ShareGPT-like format.

Important Note: This data is still in a raw format and requires careful consideration before use in training.

• Size: 13 784 rows

Data Fields

The dataset contains the following fields, mirroring the original Kajiwoto dataset structure:

• id: Unique identifier for the character.
• name: Name of the character.
• description: Character description (often empty).
• profilePhotoUri: URI of the character's profile photo (may be null).
• dominantColors: List of dominant colors in the profile photo (may be null).
• personalities: Personality tags (may be null).
• nsfw: Boolean indicating whether the character is marked as "not safe for work."
• conversations: A list of turns representing a conversation. Each turn has:
  • from: Either "human" or "gpt" indicating the speaker.
  • value: The text of the message.
• upvotes: Number of upvotes the character has received.

Data Format

The conversations field is formatted in a ShareGPT-like structure, where each turn indicates the speaker ("human" or "gpt") and the message content:

{
  "id": "M51k",
  "name": "Damian ",
  "description": "",
  "profilePhotoUri": null,
  "dominantColors": null,
  "personalities": null,
  "nsfw": false,
  "conversations": [
    {
      "from": "human",
      "value": "Still a horrific introvert?"
    },
    {
      "from": "gpt",
      "value": "I know you wanted me to find some friends, but I've already got one here in Gotham and one is more than enough"
    },
    {
      "from": "human",
      "value": "How is the friend making going "
    },
    {
      "from": "gpt",
      "value": "I know you wanted me to find some friends, but I've already got one here in Gotham and one is more than enough"
    },
    {
      "from": "human",
      "value": "You're not very good at this relax and have fun thing are you"
    },
    {
      "from": "gpt",
      "value": "I wasn't raised to have fun"
    },
   ...
  ],
  "upvotes": 0
}

Intended Uses

This dataset is intended for research and experimentation in areas such as:

• AI character development
• Dialogue generation
• Role-playing AI
• Analyzing conversational patterns

Caveats and Recommendations

Preprocessing Considerations:

• Filtering:
  • Upvotes: Consider filtering by the number of upvotes. However, be aware that profiles with fewer views might have lower upvotes, even if the quality is high.
  • NSFW: The nsfw field may not be completely reliable. Some users engage in NSFW chats with characters not marked as NSFW. Manual inspection may be required.
• Text Processing:
  • The text contains special markup that needs to be handled:
    • #scene=happy: Scene descriptions.
    • %{Howdy|G'day|Yo}: Multiple possible options for a phrase.
    • Pronouns: Requires handling for consistency.

Limitations

• The data might not be sufficiently filtered for training without further processing.
• Many short, repetitive messages may not accurately represent character behavior.
• The dataset contains a mixture of good and bad quality conversations.