import pandas as pd
from transformers import pipeline
import gradio as gr

# 載入 Hugging Face 的中文三分類情緒模型
classifier = pipeline("text-classification", model="uer/roberta-base-finetuned-dianping-chinese", tokenizer="uer/roberta-base-finetuned-dianping-chinese")

# 分析函式（接收 CSV）
def analyze_csv(file):
    df = pd.read_csv(file)

    # 檢查是否有 "text" 欄位
    if "text" not in df.columns:
        return "錯誤：CSV 檔案中必須包含 'text' 欄位。"

    # 對每一列文字做情緒分析
    results = []
    for text in df["text"]:
        result = classifier(str(text))[0]
        label = result["label"]
        score = round(result["score"], 4)

        if label == "LABEL_0":
            sentiment = "負向"
        elif label == "LABEL_1":
            sentiment = "中立"
        elif label == "LABEL_2":
            sentiment = "正向"
        else:
            sentiment = "未知"

        results.append({"label": sentiment, "score": score})

    # 加回原始 dataframe
    df["情緒判斷"] = [r["label"] for r in results]
    df["信心分數"] = [r["score"] for r in results]

    # 輸出為新的 csv 檔案
    output_file = "/tmp/output.csv"
    df.to_csv(output_file, index=False)
    return output_file

# Gradio 介面（接收檔案，回傳檔案）
gr.Interface(
    fn=analyze_csv,
    inputs=gr.File(label="上傳包含 'text' 欄位的 CSV 檔案", file_types=[".csv"]),
    outputs=gr.File(label="下載標註後的 CSV 檔案"),
    title="中文情緒分析系統（批次處理）",
    description="上傳一份 CSV，系統會針對 'text' 欄做情緒分析，並下載結果。"
).launch()