n-gram + discord

This is a CLI program that builds a n-gram language model on a character level for both users from their discord DM's.

How to use it?

C++20 and Tyrrrz/DiscordChatExporter are required.

1. Clone this repository git clone git@github.com:kiletic/ngram-discord.git
2. Enter the repository folder and export your discord DM's into a data.txt file using DiscordChatExporter CLI tool with the following command:
   dotnet DiscordChatExporter.Cli.dll export --token "INSERT_YOUR_TOKEN_HERE" -c INSERT_CHANNEL_ID -f "PlainText" -o data.txt

Note: Make sure you enter your discord token into INSERT_YOUR_TOKEN_HERE and the channel id into INSERT_CHANNEL_ID. How to get the token and the channel_id?

3. Compile main.cpp with: c++ -std=c++20 -O3 main.cpp -o main
4. Run the program: ./main INSERT_USER1_NAME INSERT_USER2_NAME

Note: Inserted names should be actual discord usernames as they appear in data.txt, and not display names. It doesn't matter in which order you write them in.

How does it work?

It parses the data.txt file for messages from both users, optionally skipping messages that contain embeddings or attachments (this can be changed by setting filter_attachments and filter_embeds to false when calling the filter_data function). Then, it builds a n-gram language model on a character level from the messages (to read more about the math check wiki). Finally, to use the model just call the generate_sentence function. Feel free to play with the value of n to get different results.