Reddit Thread Summarizer (Gradio)

Ever wanted to get the gist of a reddit thread without having to sift through memes, tldr worthy comments, and bad takes? Well, look no further!

Leverage the power of the Transformers library and Gradio to automatically generate summaries of comment threads on Reddit! The code uses state-of-the-art NLP models, such as BART, to analyze and extract the most important information from lengthy comment threads on Reddit.

This gradio app takes as input the URL of a Reddit thread and produces the following from the comments:

• Short summary
• Long summary
• Sentiment of summary
• Word cloud

The model is deployed for online use here.

Usage

The following steps are only needed if you wish to run to model locally.

First, obtain api credentials from a reddit account (instructions). After you have the relevant api secrets, create a file named api_params.toml with the following contents,

client_id = "your-client-id"
client_secret = "your-client-secret"
user_agent = "your-user_agent"

Next, install all the requirements for the code,

pip install -r requirements.txt

Finally, you can launch the app from the terminal as shown below, creating a gradio instance at a port in localhost.

python app.py

Optional

To avoid getting rate limited you need to have an access token, which you can get from having an account on huggingface.

Once you have an account you can generate a token here. After obtaining the token, add it to api_params.toml as shown.

client_id = "your-client-id"
client_secret = "your-client-secret"
user_agent = "your-user_agent"
hf_token = "your-hf-token"

TODO

• Add sentence segmentation to improve performance.
• Add sentiment analysis to output.
• Improve model inference time.