FastAPI Rest App for Toxic Comment Classification

This is a FastAPI REST app for predicting whether a given text comment is toxic or not. It uses a pre-trained DistilBERT model and tokenizer to perform the prediction.

Check out this Repo to see how I trained the DistilBERT model.

Setup

To run this app locally, follow these steps:

1. Clone the repository using git clone https://github.com/BrianMburu/Toxic-Comment-Classification-App.git.
2. Install the required packages by running pip install -r requirements.txt.
3. Run the app using uvicorn main:app --reload command and visit http://localhost:8000/docs in your web browser to interact with the API. You can also use curl as shown in the examples below.

Usage

Once the app is up and running, visit http://localhost:8000/docs in your web browser to interact with the API. You can input a text comment and choose whether to apply stemming or lemmatization before predicting its toxicity.

Input

• text: string - The input text comment to predict.
• stemmer: string - Either "snowball" or "lemmatize" to choose between applying snowball stemming or lemmatization respectively.

Output

• predictions: dictionary - A dictionary with keys representing each of the six toxicity categories and their corresponding prediction score.
• Execution Time: float - The time taken to perform the prediction, in seconds.

Examples

Input:

{
  "text": "I don't think that's a good idea",
  "stemmer": "lemmatize"
}

Output:

{
  "predictions": {
    "toxic": 0.7986,
    "severe_toxic": 0.0188,
    "obscene": 0.2733,
    "threat": 0.0183,
    "insult": 0.4133,
    "identity_hate": 0.0518
  },
  "Execution Time": 0.2174965309968684
}

Input:

{
  "text": "You are a complete loser",
  "stemmer": "snowball"
}

Output:

{
  "predictions": {
    "toxic": 0.769,
    "severe_toxic": 0.0183,
    "obscene": 0.2513,
    "threat": 0.0182,
    "insult": 0.3787,
    "identity_hate": 0.0502
  },
  "Execution Time": 0.2113154350008699
}

Example using curl command

curl -X POST "http://localhost:8000/predict" -H "Content-Type: application/json" -d '{ "text": "You are a complete loser", "stemmer": "lemmatize"}'

{"predictions":{"toxic":0.7986,"severe_toxic":0.0188,"obscene":0.2733,"threat":0.0183,"insult":0.4133,"identity_hate":0.0518},"Execution Time":0.13700330699793994}%

License

This project is licensed under the MIT License.