app.py

import streamlit as st
import pandas as pd
from transformers import AutoTokenizer, pipeline
from transformers import (
    TFAutoModelForSequenceClassification as AutoModelForSequenceClassification,
)

st.title("Detecting Toxic Tweets")

demo = """Your words are like poison. They seep into my mind and make me feel worthless."""

text = st.text_area("Input Text", demo, height=250)

model_options = {
    "DistilBERT Base Uncased (SST-2)": "distilbert-base-uncased-finetuned-sst-2-english",
    "Fine-tuned Toxicity Model": "RobCaamano/toxicity",
}
selected_model = st.selectbox("Select Model", options=list(model_options.keys()))

mod_name = model_options[selected_model]

tokenizer = AutoTokenizer.from_pretrained(mod_name)
model = AutoModelForSequenceClassification.from_pretrained(mod_name)
clf = pipeline(
    "sentiment-analysis", model=model, tokenizer=tokenizer, return_all_scores=True
)

if selected_model in ["Fine-tuned Toxicity Model"]:
    toxicity_classes = ["toxic", "severe_toxic", "obscene", "threat", "insult", "identity_hate"]
    model.config.id2label = {i: toxicity_classes[i] for i in range(model.config.num_labels)}

def get_toxicity_class(predictions, threshold=0.3):
    return {model.config.id2label[i]: pred for i, pred in enumerate(predictions) if pred >= threshold}

input = tokenizer(text, return_tensors="tf")

if st.button("Submit", type="primary"):
    results = dict(d.values() for d in clf(text)[0])
    toxic_labels = {k: results[k] for k in results.keys() if not k == "toxic"}

    tweet_portion = text[:50] + "..." if len(text) > 50 else text

    if len(toxic_labels) == 0:
        st.write("This text is not toxic.")
    else:
        df = pd.DataFrame(
            {
                "Text (portion)": [tweet_portion] * len(toxic_labels),
                "Toxicity Class": list(toxic_labels.keys()),
                "Probability": list(toxic_labels.values()),
            }
        )
        st.table(df)