chinese-text-emotion-classifier

Here's a model is fine-tuned based on another base model and features a smaller parameter size. For users who require faster inference speed, this model is a suitable choice.The actual performance test results are also not much different. Model：Chinese-Emotion-Small

此模型是基於另一個基座模型所調整的結果，擁有較小的參數規模。對於有推理速度需求的使用者，可以選擇此模型以達到更快速的性能表現，實際測試性能也相差不大。 模型：Chinese-Emotion-Small

📚 Model Introduction

This model is fine-tuned based on the joeddav/xlm-roberta-large-xnli model, specializing in Chinese text emotion analysis.
Through fine-tuning, the model can identify the following 8 emotion labels:

• Neutral tone
• Concerned tone
• Happy tone
• Angry tone
• Sad tone
• Questioning tone
• Surprised tone
• Disgusted tone

The model is applicable to various scenarios, such as customer service emotion monitoring, social media analysis, and user feedback classification.

📚 模型簡介

本模型基於joeddav/xlm-roberta-large-xnli 模型進行微調，專注於 中文語句情感分析。
通過微調，模型可以識別以下 8 種情緒標籤：

• 平淡語氣
• 關切語調
• 開心語調
• 憤怒語調
• 悲傷語調
• 疑問語調
• 驚奇語調
• 厭惡語調

該模型適用於多種場景，例如客服情緒監控、社交媒體分析以及用戶反饋分類。

🚀 Quick Start

Install Dependencies

Ensure that you have installed Hugging Face's Transformers library and PyTorch:

pip install transformers torch

###Load the Model Use the following code to load the model and tokenizer, and perform emotion classification:

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# 添加設備設定
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# 標籤映射字典
label_mapping = {
    0: "平淡語氣",
    1: "關切語調",
    2: "開心語調",
    3: "憤怒語調",
    4: "悲傷語調",
    5: "疑問語調",
    6: "驚奇語調",
    7: "厭惡語調"
}

def predict_emotion(text, model_path="Johnson8187/Chinese-Emotion"):
    # 載入模型和分詞器
    tokenizer = AutoTokenizer.from_pretrained(model_path)
    model = AutoModelForSequenceClassification.from_pretrained(model_path).to(device)  # 移動模型到設備
    
    # 將文本轉換為模型輸入格式
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True).to(device)  # 移動輸入到設備
    
    # 進行預測
    with torch.no_grad():
        outputs = model(**inputs)
    
    # 取得預測結果
    predicted_class = torch.argmax(outputs.logits).item()
    predicted_emotion = label_mapping[predicted_class]
    
    return predicted_emotion

if __name__ == "__main__":
    # 使用範例
    test_texts = [
        "雖然我努力了很久，但似乎總是做不到，我感到自己一無是處。",
        "你說的那些話真的讓我很困惑，完全不知道該怎麼反應。",
        "這世界真的是無情，為什麼每次都要給我這樣的考驗？",
        "有時候，我只希望能有一點安靜，不要再聽到這些無聊的話題。",
        "每次想起那段過去，我的心還是會痛，真的無法釋懷。",
        "我從來沒有想過會有這麼大的改變，現在我覺得自己完全失控了。",
        "我完全沒想到你會這麼做，這讓我驚訝到無法言喻。",
        "我知道我應該更堅強，但有些時候，這種情緒真的讓我快要崩潰了。"
    ]

    for text in test_texts:
        emotion = predict_emotion(text)
        print(f"文本: {text}")
        print(f"預測情緒: {emotion}\n")

🚀 快速開始

安裝依賴

請確保安裝了 Hugging Face 的 Transformers 庫和 PyTorch：

pip install transformers torch

加載模型

使用以下代碼加載模型和分詞器，並進行情感分類：

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# 添加設備設定
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# 標籤映射字典
label_mapping = {
    0: "平淡語氣",
    1: "關切語調",
    2: "開心語調",
    3: "憤怒語調",
    4: "悲傷語調",
    5: "疑問語調",
    6: "驚奇語調",
    7: "厭惡語調"
}

def predict_emotion(text, model_path="Johnson8187/Chinese-Emotion"):
    # 載入模型和分詞器
    tokenizer = AutoTokenizer.from_pretrained(model_path)
    model = AutoModelForSequenceClassification.from_pretrained(model_path).to(device)  # 移動模型到設備
    
    # 將文本轉換為模型輸入格式
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True).to(device)  # 移動輸入到設備
    
    # 進行預測
    with torch.no_grad():
        outputs = model(**inputs)
    
    # 取得預測結果
    predicted_class = torch.argmax(outputs.logits).item()
    predicted_emotion = label_mapping[predicted_class]
    
    return predicted_emotion

if __name__ == "__main__":
    # 使用範例
    test_texts = [
        "雖然我努力了很久，但似乎總是做不到，我感到自己一無是處。",
        "你說的那些話真的讓我很困惑，完全不知道該怎麼反應。",
        "這世界真的是無情，為什麼每次都要給我這樣的考驗？",
        "有時候，我只希望能有一點安靜，不要再聽到這些無聊的話題。",
        "每次想起那段過去，我的心還是會痛，真的無法釋懷。",
        "我從來沒有想過會有這麼大的改變，現在我覺得自己完全失控了。",
        "我完全沒想到你會這麼做，這讓我驚訝到無法言喻。",
        "我知道我應該更堅強，但有些時候，這種情緒真的讓我快要崩潰了。"
    ]

    for text in test_texts:
        emotion = predict_emotion(text)
        print(f"文本: {text}")
        print(f"預測情緒: {emotion}\n")

Dataset

• The fine-tuning dataset consists of 4,000 annotated Traditional Chinese emotion samples, covering various emotion categories to ensure the model's generalization capability in emotion classification.
• Johnson8187/Chinese_Multi-Emotion_Dialogue_Dataset

數據集

• 微調數據來自4000個自行標註的高質量繁體中文情感語句數據，覆蓋了多種情緒類別，確保模型在情感分類上的泛化能力。
• Johnson8187/Chinese_Multi-Emotion_Dialogue_Dataset

🌟 Contact and Feedback If you encounter any issues while using this model, please contact:

Email: fable8043@gmail.com Hugging Face Project Page: chinese-text-emotion-classifier

🌟 聯繫與反饋

如果您在使用該模型時有任何問題，請聯繫：

• 郵箱：fable8043@gmail.com
• Hugging Face 項目頁面：chinese-text-emotion-classifier

@misc{lin2024chinese_emotion,
  author = {Lin, Cheng-Yen},
  title = {Chinese-Emotion: Fine-grained Emotion Classification Model for Traditional Chinese},
  year = {2024},
  publisher = {Hugging Face},
  journal = {Hugging Face repository},
  howpublished = {\url{https://huggingface.co/Johnson8187/Chinese-Emotion}}
}