app.py

import gradio as gr
import torch
import torch.nn.functional as F
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline, DistilBertTokenizer, DistilBertForSequenceClassification

# ---------------- Original Sarcasm + Sentiment Models ----------------
sarcasm_model = AutoModelForSequenceClassification.from_pretrained("dnzblgn/Sarcasm-Detection-Customer-Reviews")
sarcasm_tokenizer = AutoTokenizer.from_pretrained("dnzblgn/Sarcasm-Detection-Customer-Reviews", use_fast=False)

sentiment_model = AutoModelForSequenceClassification.from_pretrained("dnzblgn/Sentiment-Analysis-Customer-Reviews")
sentiment_tokenizer = AutoTokenizer.from_pretrained("dnzblgn/Sentiment-Analysis-Customer-Reviews", use_fast=False)

def analyze_sentiment(sentence):
    inputs = sentiment_tokenizer(sentence, return_tensors="pt", truncation=True, padding=True, max_length=512)
    with torch.no_grad():
        outputs = sentiment_model(**inputs)
    logits = outputs.logits
    predicted_class = torch.argmax(logits, dim=-1).item()
    sentiment_mapping = {1: "Negative", 0: "Positive"}
    return sentiment_mapping[predicted_class]

def detect_sarcasm(sentence):
    inputs = sarcasm_tokenizer(sentence, return_tensors="pt", truncation=True, padding=True, max_length=512)
    with torch.no_grad():
        outputs = sarcasm_model(**inputs)
    logits = outputs.logits
    predicted_class = torch.argmax(logits, dim=-1).item()
    return "Sarcasm" if predicted_class == 1 else "Not Sarcasm"

def process_text_pipeline(text):
    sentences = text.split("\n")
    processed_sentences = []

    for sentence in sentences:
        sentence = sentence.strip()
        if not sentence:
            continue

        sentiment = analyze_sentiment(sentence)
        if sentiment == "Negative":
            processed_sentences.append(f"❌ '{sentence}' -> Sentiment: Negative")
        else:
            sarcasm_result = detect_sarcasm(sentence)
            if sarcasm_result == "Sarcasm":
                processed_sentences.append(f"⚠️ '{sentence}' -> Sentiment: Negative (Sarcastic Positive)")
            else:
                processed_sentences.append(f"✅ '{sentence}' -> Sentiment: Positive")

    return "\n".join(processed_sentences)

# ---------------- Additional Sentiment Models (No Sarcasm) ----------------
# Pre-load tokenizers + models for safety
additional_models = {
    "siebert/sentiment-roberta-large-english": {
        "tokenizer": AutoTokenizer.from_pretrained("siebert/sentiment-roberta-large-english"),
        "model": AutoModelForSequenceClassification.from_pretrained("siebert/sentiment-roberta-large-english")
    },
    "assemblyai/bert-large-uncased-sst2": {
        "tokenizer": AutoTokenizer.from_pretrained("assemblyai/bert-large-uncased-sst2"),
        "model": AutoModelForSequenceClassification.from_pretrained("assemblyai/bert-large-uncased-sst2")
    },
    "j-hartmann/sentiment-roberta-large-english-3-classes": {
        "tokenizer": AutoTokenizer.from_pretrained("j-hartmann/sentiment-roberta-large-english-3-classes"),
        "model": AutoModelForSequenceClassification.from_pretrained("j-hartmann/sentiment-roberta-large-english-3-classes")
    },
    "cardiffnlp/twitter-xlm-roberta-base-sentiment": {
        "tokenizer": AutoTokenizer.from_pretrained("cardiffnlp/twitter-xlm-roberta-base-sentiment"),
        "model": AutoModelForSequenceClassification.from_pretrained("cardiffnlp/twitter-xlm-roberta-base-sentiment")
    },
    "sohan-ai/sentiment-analysis-model-amazon-reviews": {
        "tokenizer": DistilBertTokenizer.from_pretrained("distilbert-base-uncased"),
        "model": DistilBertForSequenceClassification.from_pretrained("sohan-ai/sentiment-analysis-model-amazon-reviews")
    }
}

def run_sentiment_with_selected_model(text, model_name):
    model_info = additional_models[model_name]
    tokenizer = model_info["tokenizer"]
    model = model_info["model"]

    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
    with torch.no_grad():
        outputs = model(**inputs)

    logits = outputs.logits
    probs = F.softmax(logits, dim=-1)
    pred = torch.argmax(probs, dim=-1).item()

    # Custom label mapping
    label_map = {
        "assemblyai/bert-large-uncased-sst2": {0: "Negative", 1: "Positive"},
        "sohan-ai/sentiment-analysis-model-amazon-reviews": {0: "Negative", 1: "Positive"},
    }

    if model_name in label_map:
        label = label_map[model_name][pred]
    elif model.config.id2label:
        label = model.config.id2label.get(pred, f"LABEL_{pred}")
    else:
        label = f"LABEL_{pred}"

    emoji = "✅" if "positive" in label.lower() else "❌" if "negative" in label.lower() else "⚠️"
    
    # Add confidence score
    confidence = probs[0][pred].item() * 100
    return f"{emoji} '{text}' -> {label} ({confidence:.1f}%)"

# ---------------- Gradio UI ----------------
background_css = """
.gradio-container {
    background-image: url('https://huggingface.co/spaces/dnzblgn/Sarcasm_Detection/resolve/main/image.png');
    background-size: cover;
    background-position: center;
    color: white;
}
.gr-input, .gr-textbox {
    background-color: rgba(255, 255, 255, 0.3) !important;
    border-radius: 10px;
    padding: 10px;
    color: black !important;
}
h1, h2, p {
    text-shadow: 1px 1px 2px rgba(0, 0, 0, 0.8);
}
"""

with gr.Blocks(css=background_css) as interface:
    gr.Markdown(
        """
        <h1 style='text-align: center; font-size: 36px;'>🌟 Sentiment Analysis Powered by Sarcasm Detection 🌟</h1>
        <p style='text-align: center; font-size: 18px;'>Analyze the sentiment of customer reviews and detect sarcasm in positive reviews.</p>
        """
    )

    with gr.Tab("Text Input"):
        with gr.Row():
            text_input = gr.Textbox(lines=10, label="Enter Sentences", placeholder="Enter one or more sentences, each on a new line.")
            result_output = gr.Textbox(label="Results", lines=10, interactive=False)
        analyze_button = gr.Button("🔍 Analyze")
        analyze_button.click(process_text_pipeline, inputs=text_input, outputs=result_output)

    with gr.Tab("Upload Text File"):
        file_input = gr.File(label="Upload Text File")
        file_output = gr.Textbox(label="Results", lines=10, interactive=False)

        def process_file(file):
            text = file.read().decode("utf-8")
            return process_text_pipeline(text)

        file_input.change(process_file, inputs=file_input, outputs=file_output)

    with gr.Tab("Try Other Sentiment Models"):
        with gr.Row():
            other_model_selector = gr.Dropdown(
                choices=list(additional_models.keys()),
                label="Choose a Sentiment Model"
            )
        with gr.Row():
            model_text_input = gr.Textbox(lines=5, label="Enter Sentence")
            model_result_output = gr.Textbox(label="Sentiment", lines=3, interactive=False)

        run_model_btn = gr.Button("Run")
        run_model_btn.click(run_sentiment_with_selected_model, inputs=[model_text_input, other_model_selector], outputs=model_result_output)

# ---------------- Run App ----------------
if __name__ == "__main__":
    interface.launch()