CourseBench Feedback Refinement System

This project is a FastAPI application that uses a GPT generator to refine the language used in forum messages. The system filters out inappropriate words or phrases, ensuring a positive and constructive environment for university students to discuss and comment on their courses.

The refinement is following code below in example/system.txt

• Leaving well-structured and informative messages unaltered
• Preserving main points of critical and emotive messages while removing offensive language
• Transforming highly emotional messages into more objective ones
• Restructuring incoherent or disorganized messages to improve clarity
• If encountering unfamiliar words or phrases, feel free to disregard them
• For messages containing sarcastic or mocking, maintain the original content as long as it does not include insults directed at a professor's character
• Negative comments about courses or teaching methods are permissible, but refrain from using offensive words regarding a professor's personality.

Features

• Exposes a POST endpoint at /v1/generate which accepts parameters for content and title.
• Uses a GPT generator to generate responses based on the provided content and title.
• Handles exceptions during the generation of messages, returning a HTTPException with status code 500.

Installation

The project requires Python 3.10.12 and the dependencies listed in the requirements.txt file. You can install these dependencies using pip:

pip install -r requirements.txt

Usage

To start the server, you need to set the "OPENAI_API_BASE" and "OPENAI_API_KEY" environment variables. You can do this in the terminal before running the server:

export OPENAI_API_BASE='your_openai_api_base'
export OPENAI_API_KEY='your_openai_api_key'
uvicorn app:app --port 80 --host 0.0.0.0

To use the /v1/generate endpoint, send a POST request with a JSON body containing the title and content parameters. For example:

{
  "title": "Course Feedback",
  "content": "The course was challenging but rewarding."
}

Then, the response could be:

{
  "title": "AAAA",
  "content": "BBBB",
  "reason": "XXXX"
}

Docker

A Dockerfile is provided for containerization. To build and run the Docker container, you need to pass the "OPENAI_API_BASE" and "OPENAI_API_KEY" as environment variables. Use the following commands:

docker build -t course-feedback-refinement-system .
docker run -p 80:80 -e OPENAI_API_BASE='your_openai_api_base' -e OPENAI_API_KEY='your_openai_api_key' course-feedback-refinement-system

Testing

Test cases are provided in the testcases directory. Each test case is a json file containing a message that needs to be refined.

You need to install Gradio first:

pip install gradio

Then, run test script

python test.py

The interactive interface will be shown in the browser, by default, it will be shown in http://127.0.0.1:7860/. You can click the "Generate" button to generate the refined message. The refined message will be shown in the "Output" box.

Contributing

Contributions are welcome. Please make sure to update the test cases as appropriate when making a change.

License

This project is licensed under the terms of the MIT license.

Acknowledge

Documentation is written with GPT-4.