标签:

• 文本分类
• 多标签
• GoEmotions情感数据集
• transformers库
• huggingface平台 许可证: apache-2.0 库名称: transformers 语言:
• 英文 评估指标:
• 准确率
• F1值 基础模型:
• google-bert/bert-base-uncased 管道标签: 文本分类

🔥 基于GoEmotions数据集微调的BERT模型

📖 模型概述

本模型是在GoEmotions情感数据集上微调的BERT模型(bert-base-uncased)，专为多标签情感分类任务设计，可对输入文本预测多个情感标签。

📊 性能指标

📂 使用示例

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# 加载模型和分词器
model_name = "codewithdark/bert-Gomotions"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# 情感标签列表（根据数据集调整）
emotion_labels = [
    "钦佩", "有趣", "愤怒", "烦恼", "赞同", "关心", "困惑",
    "好奇", "渴望", "失望", "反对", "厌恶", "尴尬",
    "兴奋", "恐惧", "感激", "悲伤", "快乐", "爱", "紧张", "乐观",
    "骄傲", "领悟", "解脱", "悔恨", "伤心", "惊讶", "中性"
]

# 示例文本
text = "我今天太开心了！"
inputs = tokenizer(text, return_tensors="pt")

# 预测
with torch.no_grad():
    outputs = model(**inputs)
    probs = torch.sigmoid(outputs.logits).squeeze(0)  # 将logits转为概率值

# 获取前5个预测结果
top5_indices = torch.argsort(probs, descending=True)[:5]  # 获取前5个标签索引
top5_labels = [emotion_labels[i] for i in top5_indices]
top5_probs = [probs[i].item() for i in top5_indices]

# 打印结果
print("预测情感TOP5:")
for label, prob in zip(top5_labels, top5_probs):
    print(f"{label}: {prob:.4f}")

'''
输出示例:
预测情感TOP5:
快乐: 0.9478
爱: 0.7854
乐观: 0.6342
钦佩: 0.5678
兴奋: 0.5231
'''

🏋️‍♂️ 训练细节

• 基础模型: bert-base-uncased
• 数据集: GoEmotions
• 优化器: AdamW
• 损失函数: BCEWithLogitsLoss（多标签分类的二值交叉熵）
• 批大小: 16
• 训练轮次: 3
• 评估指标: 准确率、F1值、汉明损失

📌 Hugging Face快速调用

from transformers import pipeline

classifier = pipeline("text-classification", model="codewithdark/bert-Gomotions", top_k=None)
classifier("我对这次旅行太期待了！")

🛠️ 引用方式

如需使用本模型，请引用：

@misc{your_model,
  author = {codewithdark},
  title = {基于GoEmotions微调的BERT模型},
  year = {2025},
  url = {https://huggingface.co/codewithdark/bert-Gomotions}
}