app.py

# https://aistudio.google.com/app/u/1/apikey <--- Get Your Google GenAI API Key Here.

import os
import time
import pandas as pd
import streamlit as st
from google import genai
import plotly.express as px
from dotenv import load_dotenv

# Load environment
load_dotenv()
api_key = os.getenv("GOOGLE_GENAI_API_KEY")
client = genai.Client(api_key = api_key)

# Page config
st.set_page_config(page_title = "GenAI-TopicModeling")

# Title section
st.markdown("""
    <style>
        .centered-title {
            text-align: center;
            font-size: 36px;
            font-weight: bold;
        }
    </style>
    <h1 class="centered-title">
        <span style="color:#00843D;">GenAI Topic</span> 
        <span style="color:black;">Modeling Tool.</span>
    </h1>
    <br>
    """, unsafe_allow_html = True)

# Upload CSV
uploaded_file = st.file_uploader(
    "Upload CSV for analysis. (NOTE: each row representing a new document)", type = ["csv"])

if uploaded_file:
    df = pd.read_csv(uploaded_file)
    st.write("Data Preview:")
    st.dataframe(df.head(), use_container_width = True)

    # Column selection
    text_column = st.selectbox("Select the column for analysis:", df.columns)

    # Category selection
    category_mode = st.radio(
        "Would you like to work with the default categories or provide custom categories?",
        ["Default Categories", "Custom Categories"],
        help = "Default categories: pro-police, pro-protester, anti-police, anti-protester, neutral")

    if category_mode == "Custom Categories":
        custom_input = st.text_area(
            "Enter your custom categories (comma-separated), THEN click anywhere outside the text box:",
            placeholder = "e.g. category1, category2, category3")
        categories = [c.strip() for c in custom_input.split(",") if c.strip()]
    else:
        categories = ["pro-police", "pro-protester", "anti-police", "anti-protester", "neutral"]

    # Check input validity
    if not categories:
        st.warning("Please enter at least one valid category.")
        st.stop()

    if df[text_column].isnull().any():
        st.error("Selected column contains missing values.")
        st.stop()

    # Begin Analysis
    st.markdown("### Document Classification")
    with st.spinner("Classifying documents using Gemini 2.0 Flash-Lite..."):
        texts = df[text_column].astype(str).tolist()
        processed_labels = []
        batch_size = 30

        for batch_start in range(0, len(texts), batch_size):
            batch_texts = texts[batch_start:batch_start + batch_size]
            joined_texts = "\n\n".join(
                [f"{i + 1}. {t}" for i, t in enumerate(batch_texts)])

            prompt = (
                f"You are a classifier. For each of the following {len(batch_texts)} documents, "
                f"assign **one label** from this list: {categories}.\n\n"
                f"Documents:\n{joined_texts}\n\n"
                f"Return a numbered list of {len(batch_texts)} labels in the same order. "
                f"Example output format:\n1. label1\n2. label2\n...\n{len(batch_texts)}. label{len(batch_texts)}")

            try:
                response = client.models.generate_content(
                    model = "gemini-2.0-flash-lite",
                    contents = prompt)

                response_text = response.text.strip()
                batch_labels = []

                for line in response_text.splitlines():
                    if "." in line:
                        _, label = line.split(".", 1)
                        batch_labels.append(label.strip())

                if len(batch_labels) != len(batch_texts):
                    raise ValueError(
                        f"Expected {len(batch_texts)} labels, got {len(batch_labels)}")

                processed_labels.extend(batch_labels)
                time.sleep(1)

            except Exception as e:
                st.error(f"Error processing batch {batch_start}-{batch_start + batch_size}: {str(e)}")
                st.stop()

        # Combine results
        df["Topics"] = processed_labels
        topic_counts = df["Topics"].value_counts().reset_index()
        topic_counts.columns = ["Topic", "Frequency"]

    st.success("Analysis completed successfully!")

    # Results Section
    st.divider()
    st.markdown("""
        <style>
            .centered-result {
                text-align: center;
                font-size: 36px;
                font-weight: bold;
            }
        </style>
        <h1 class="centered-result">
            <span style="color:#00843D;">Analysis</span> 
            <span style="color:black;">Results.</span>
        </h1>
        <br>
        """, unsafe_allow_html = True)

    # Plot
    st.markdown("### Topic Frequency Plot")
    fig = px.bar(
        topic_counts,
        x = "Topic", y = "Frequency",
        color = "Topic",
        title = "Distribution of Topics",
        template = "plotly_white")

    st.plotly_chart(fig, use_container_width = True)

    # Document Search
    st.markdown(f"### Search `{text_column}` Content")
    search_term = st.text_input("Search in document texts:").strip().lower()

    if search_term:
        filtered_df = df[df[text_column].str.lower().str.contains(search_term)]
    else:
        filtered_df = df

    st.markdown("### Labeled Documents")
    st.dataframe(filtered_df[[text_column, "Topics"]], use_container_width = True, height = 300)