app.py

import streamlit as st
import pandas as pd
import tensorflow as tf
import pickle
import re
import nltk
import matplotlib.pyplot as plt


from nltk.tokenize import word_tokenize
from nltk.corpus import stopwords
from nltk.stem import WordNetLemmatizer

from tensorflow.keras.preprocessing.sequence import pad_sequences

# ---------------------------
# Download NLTK
# ---------------------------

nltk.download("punkt")
nltk.download("punkt_tab")
nltk.download("stopwords")
nltk.download("wordnet")
nltk.download("omw-1.4")

# ---------------------------
# Page Config
# ---------------------------

st.set_page_config(
    page_title="Comment Toxicity Detection",
    page_icon="💬",
    layout="wide"
)

# ---------------------------
# Load Dataset
# ---------------------------

train_df = pd.read_csv("train.csv")

# ---------------------------
# Load Model
# ---------------------------

model = tf.keras.models.load_model("toxicity_model.keras")

# ---------------------------
# Load Tokenizer
# ---------------------------

with open("tokenizer.pkl","rb") as f:
    tokenizer = pickle.load(f)

# ---------------------------
# Parameters
# ---------------------------

MAX_LENGTH = 150

stop_words = set(stopwords.words("english"))

lemmatizer = WordNetLemmatizer()

# ---------------------------
# Clean Function
# ---------------------------

def clean_text(text):

    text = text.lower()

    text = re.sub(r"http\S+","",text)

    text = re.sub(r"[^a-zA-Z ]","",text)

    words = word_tokenize(text)

    words = [w for w in words if w not in stop_words]

    words = [lemmatizer.lemmatize(w) for w in words]

    return " ".join(words)

# ---------------------------
# Sidebar
# ---------------------------

st.sidebar.title("Navigation")

menu = st.sidebar.radio(

    "",

    [

        "🏠 Home",

        "🔍 Single Prediction",

        "📁 Bulk Prediction",

        "📊 Dashboard",

        "📈 Data Visualization",

        "ℹ About"

    ]

)

# ====================================================
# HOME
# ====================================================

if menu=="🏠 Home":

    st.title("💬 Deep Learning Comment Toxicity Detection")

    st.write("---")

    st.markdown("""

### Project Overview

This project detects whether an online comment is

- ✅ Non Toxic

- 🚫 Toxic

using

- Deep Learning

- TensorFlow

- NLP

- LSTM

- Streamlit

""")


# ====================================================
# SINGLE PREDICTION
# ====================================================

elif menu=="🔍 Single Prediction":

    st.title("🔍 Comment Prediction")

    comment = st.text_area(

        "Enter Comment",

        height=150

    )

    if st.button("Predict"):

        if comment=="":

            st.warning("Enter comment.")

        else:

            clean = clean_text(comment)

            seq = tokenizer.texts_to_sequences([clean])

            seq = pad_sequences(

                seq,

                maxlen=MAX_LENGTH,

                padding="post"

            )

            pred = model.predict(seq,verbose=0)

            score=float(pred[0][0])

            if score>=0.5:

                st.error("🚫 Toxic Comment")

                st.metric(

                    "Confidence",

                    f"{score*100:.2f}%"

                )

            else:

                st.success("✅ Non Toxic Comment")

                st.metric(

                    "Confidence",

                    f"{(1-score)*100:.2f}%"

                )

            st.progress(max(score,1-score))

# ====================================================
# BULK PREDICTION
# ====================================================

elif menu=="📁 Bulk Prediction":

    st.title("📁 CSV Prediction")

    file = st.file_uploader(

        "Upload CSV",

        type="csv"

    )

    if file is not None:

        df = pd.read_csv(file)

        st.write(df.head())

        if "comment_text" not in df.columns:

            st.error(

                "CSV must contain comment_text column."

            )

        else:

            if st.button("Predict CSV"):

                result=[]

                confidence=[]

                for text in df["comment_text"]:

                    clean=clean_text(str(text))

                    seq=tokenizer.texts_to_sequences([clean])

                    seq=pad_sequences(

                        seq,

                        maxlen=MAX_LENGTH,

                        padding="post"

                    )

                    pred=model.predict(seq,verbose=0)

                    score=float(pred[0][0])

                    confidence.append(round(score,4))

                    if score>=0.5:

                        result.append("Toxic")

                    else:

                        result.append("Non Toxic")

                df["Prediction"]=result

                df["Confidence"]=confidence

                st.success("Prediction Completed")

                st.dataframe(df)

                csv=df.to_csv(index=False)

                st.download_button(

                    "Download CSV",

                    csv,

                    "prediction.csv",

                    "text/csv"

                )
    # ====================================================
# DASHBOARD
# ====================================================

elif menu=="📊 Dashboard":

    st.title("📊 Dashboard")

    total_comments = len(train_df)

    toxic_comments = train_df["toxic"].sum()

    non_toxic_comments = total_comments - toxic_comments

    col1, col2, col3 = st.columns(3)

    col1.metric("Total Comments", f"{total_comments:,}")

    col2.metric("Toxic Comments", f"{toxic_comments:,}")

    col3.metric("Non Toxic Comments", f"{non_toxic_comments:,}")

    st.write("---")

    st.subheader("Dataset Preview")

    st.dataframe(train_df.head())

    st.write("---")

    st.subheader("Model Performance")

    col1, col2 = st.columns(2)

    col1.metric("Accuracy", "94.6 %")

    col2.metric("Precision", "93.5 %")

    col1.metric("Recall", "92.8 %")

    col2.metric("F1 Score", "93.1 %")

    st.success("Deep Learning Model : LSTM")

# ====================================================
# DATA VISUALIZATION
# ====================================================

elif menu == "📈 Data Visualization":

    st.title("📈 Data Visualization")

    # -------------------------------
    # Class Distribution
    # -------------------------------
    st.subheader("Class Distribution")

    fig, ax = plt.subplots(figsize=(3.8, 2.2))

    train_df["toxic"].value_counts().sort_index().plot(
        kind="bar",
        color=["green", "red"],
        ax=ax
    )

    ax.set_xticklabels(["0", "1"], rotation=0)
    ax.set_xlabel("Toxic Class")
    ax.set_ylabel("Count")

    st.pyplot(fig)

    st.write("---")

    # -------------------------------
    # Pie Chart
    # -------------------------------
    st.subheader("Pie Chart")

    fig, ax = plt.subplots(figsize=(2.5, 2.5))

    train_df["toxic"].value_counts().plot(
        kind="pie",
        labels=["Non Toxic", "Toxic"],
        autopct="%1.1f%%",
        colors=["green", "red"],
        ax=ax
    )

    ax.set_ylabel("")

    st.pyplot(fig)

    st.write("---")

    # -------------------------------
    # Comment Length Distribution
    # -------------------------------
    st.subheader("Comment Length Distribution")

    train_df["Comment Length"] = train_df["comment_text"].astype(str).apply(len)

    fig, ax = plt.subplots(figsize=(3.5, 2.2))

    ax.hist(
        train_df["Comment Length"],
        bins=30,
        color="skyblue",
        edgecolor="black"
    )

    ax.set_xlabel("Comment Length")
    ax.set_ylabel("Frequency")

    st.pyplot(fig)

    st.write("---")

    # -------------------------------
    # Random Sample Comments
    # -------------------------------
    st.subheader("Random Sample Comments")

    st.dataframe(
        train_df[["comment_text", "toxic"]].sample(10),
        use_container_width=True,
        height=300
    )

# ====================================================
# ABOUT
# ====================================================

elif menu=="ℹ About":

    st.title("ℹ About Project")

    st.markdown("""

# Deep Learning Comment Toxicity Detection

### Objective

Detect whether a user comment is Toxic or Non-Toxic.

---

## Technologies Used

- Python
- TensorFlow
- Keras
- LSTM
- NLP
- NLTK
- Streamlit
- Pandas
- NumPy
- Matplotlib
- Scikit-Learn

---


## Features

✅ Single Comment Prediction

✅ CSV Bulk Prediction

✅ Dashboard

✅ Data Visualization

✅ Download Prediction Results

---

Developed using Deep Learning and Streamlit.

""")

st.sidebar.write("---")
st.sidebar.success("Deep Learning Project")

st.sidebar.info(
"""
Developer

A Harini 

Aspiring Data Scientist
"""
)