RedBot

RedBot is a chatbot trained on Reddit comments dataset using a transformer model using Tensorflow framework. Also tkinter has been used to create a GUI which allows the user to interact with the chatbot.

Installation and Setup

Fork the repo and clone it.

git clone https://github.com/Frostday/RedBot.git

Install the necessary packages required to run the code

pip install -r requirements.txt

Run the following command

python -u chatbot_gui.py

File Structure

data folder contains the preprocessed data which was made using chatbot_database.py and create_training_data.py scripts.
saved models folder contains the pretrained models and the colab notebook shows how these models were trained.
chatbot_gui.py is the file which contains the code for GUI and model.py contains the code for reconstructing the pre-trained model so it can be used inside chatbot_gui.py file.

Dataset

The raw dataset can be found here - http://files.pushshift.io/reddit/comments/. This dataset contains millions of reddit comments and their responses from the year 2016 to 2020 from many different subreddits. I recommend downloading .xz files as they contain the same data as .bz2 files but use higher compression.

Model

I used the model which was proposed in one of Google's research papers - "Attention is all you need". A basic implementation of this model can be found here - https://blog.tensorflow.org/2019/05/transformer-chatbot-tutorial-with-tensorflow-2.html.

Results

200 epochs and 100,000 observations

Loss	Accuracy
0.6568	0.2676