app.py

import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from transformers import BertTokenizer, BertForSequenceClassification
import gradio as gr
import os
import zipfile

# --------- Sentiment Model (Binary, expanded to 3 classes) ---------
sentiment_model_name = "uer/roberta-base-finetuned-jd-binary-chinese"
sentiment_tokenizer = AutoTokenizer.from_pretrained(sentiment_model_name)
sentiment_model = AutoModelForSequenceClassification.from_pretrained(sentiment_model_name)
sentiment_model.eval()
if not os.path.exists("result"):
    with zipfile.ZipFile("model_output.zip", "r") as zip_ref:
        zip_ref.extractall(".")


# 加载你的多标签分类模型
label_tokenizer = AutoTokenizer.from_pretrained("result")
label_model = AutoModelForSequenceClassification.from_pretrained("result", use_safetensors=True)
label_model.eval()

# 多标签类别
label_map = {
    0: "Landscape & Culture",
    1: "Service & Facilities",
    2: "Experience & Atmosphere",
    3: "Transportation Accessibility",
    4: "Interactive Activities",
    5: "Price & Consumption"
}
threshold = 0.5


# --------- Multi-label Classification Model (Your model) ---------
label_dir = "./result"
label_tokenizer = BertTokenizer.from_pretrained(label_dir)
label_model = BertForSequenceClassification.from_pretrained(label_dir)
label_model.eval()

# Label categories
label_map = {
    0: "Landscape & Culture",
    1: "Service & Facilities",
    2: "Experience & Atmosphere",
    3: "Transportation Accessibility",
    4: "Interactive Activities",
    5: "Price & Consumption"
}
threshold = 0.5

# --------- Inference Function ---------
def analyze(text):
    if not text.strip():
        return "Please enter a valid comment.", "Please enter a valid comment."

    # --- Sentiment Analysis ---
    sent_inputs = sentiment_tokenizer(text, return_tensors='pt', truncation=True, padding=True, max_length=128)
    with torch.no_grad():
        sent_outputs = sentiment_model(**sent_inputs)
        probs = torch.softmax(sent_outputs.logits, dim=1).squeeze().tolist()

    pos_prob, neg_prob = probs[1], probs[0]
    if abs(pos_prob - neg_prob) < 0.2:
        sentiment_label = "Neutral"
    elif pos_prob > neg_prob:
        sentiment_label = "Positive"
    else:
        sentiment_label = "Negative"

    sentiment_result = (
        f"Prediction: {sentiment_label}\n\n"
        f"Sentiment Scores:\n"
        f"Positive: {pos_prob:.2f}\n"
        f"Neutral: {1 - abs(pos_prob - neg_prob):.2f} \n"
        f"Negative: {neg_prob:.2f}"
    )

    # --- Label Prediction ---
    label_inputs = label_tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=128)
    with torch.no_grad():
        label_outputs = label_model(**label_inputs)
        logits = label_outputs.logits
        probs = torch.sigmoid(logits).squeeze().tolist()

    if isinstance(probs, float):
        probs = [probs]

    selected_labels = [label_map[i] for i, p in enumerate(probs) if p >= threshold]
    if selected_labels:
        label_result = "Predicted Tags:\n" + "\n".join([f"{label_map[i]} ({probs[i]:.2f})" for i in range(len(probs)) if probs[i] >= threshold])
    else:
        label_result = "No confident labels identified by the model."

    return sentiment_result, label_result

# --------- Gradio Web UI ---------
with gr.Blocks(title="Sentiment + Tag Analysis System") as demo:
    gr.Markdown("## 🌟  Comment Analyzer")
    gr.Markdown(
        "This tool analyzes **Tourist comment data** using deep learning models. "
        "It predicts both **sentiment polarity** (Positive / Neutral / Negative) and **semantic category tags** (6 themes)."
    )

    with gr.Row():
        with gr.Column():
            input_box = gr.Textbox(label="Enter a review", placeholder="e.g., The park is peaceful and the staff are friendly...", lines=4)
            submit_btn = gr.Button("🔍 Analyze")
        with gr.Column():
            sentiment_output = gr.Textbox(label="Sentiment Result", lines=6)
            label_output = gr.Textbox(label="Tag Classification Result", lines=6)

    submit_btn.click(fn=analyze, inputs=input_box, outputs=[sentiment_output, label_output])

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