license: cc-by-sa-4.0 datasets:

• Kostya165/ru_emotion_dvach language:
• ru metrics:
• accuracy base_model:
• cointegrated/rubert-tiny2 pipeline_tag: text-classification tags:
• russian
• emotion
• sentiment
• sentiment-analisys
• emotion-analisys
• emotion-classification
• emotion-detection
• rubert
• rubert-tiny

rubert_tiny2_russian_emotion_sentiment

模型说明

rubert_tiny2_russian_emotion_sentiment是基于轻量级模型cointegrated/rubert-tiny2微调的情感分类模型，用于识别俄语文本中的五种情绪：

• 0: 攻击性（агрессия）
• 1: 焦虑（тревожность）
• 2: 中性（нейтральное состояние）
• 3: 积极（позитив）
• 4: 讽刺（сарказм）

验证集表现

各类别准确率：

• 攻击性 (0): 0.9120
• 焦虑 (1): 0.9462
• 中性 (2): 0.8663
• 积极 (3): 0.8884
• 讽刺 (4): 0.8426

使用方式

pip install transformers torch

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# 加载模型和分词器
MODEL_ID = "Kostya165/rubert_tiny2_russian_emotion_sentiment"
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model     = AutoModelForSequenceClassification.from_pretrained(MODEL_ID)
model.eval()

texts = [
    "今天真是美好的一天！",
    "这一切都让我火冒三丈。"
]

# 文本编码
enc = tokenizer(texts, padding=True, truncation=True, max_length=128, return_tensors="pt")
with torch.no_grad():
    logits = model(**enc).logits
    preds = logits.argmax(dim=-1).tolist()

# 将预测ID转换为标签
id2label = model.config.id2label
labels = [id2label[p] for p in preds]
print(labels)  # 输出示例: ['positive', 'aggression']

训练细节

• 基础模型: cointegrated/rubert-tiny2
• 训练数据: Kostya165/ru_emotion_dvach
• 训练轮次: 2
• 批大小: 32
• 学习率: 1e-5
• 混合精度: FP16
• 正则化: Dropout 0.1, weight_decay 0.01, warmup_ratio 0.1

依赖环境

• transformers>=4.30.0
• torch>=1.10.0
• datasets
• evaluate

许可协议

CC-BY-SA 4.0.

引用方式

@article{rubert_tiny2_russian_emotion_sentiment,
  title   = {基于RuBERT-tiny2的俄语情感分类模型},
  author  = {Kostya165},
  year    = {2024},
  howpublished = {\url{https://huggingface.co/Kostya165/rubert_tiny2_russian_emotion_sentiment}}
}

英文版说明

rubert_tiny2_russian_emotion_sentiment

模型概述

本模型是在轻量级cointegrated/rubert-tiny2基础上微调的俄语情感分类器，可识别五种情绪：

• 0: 攻击性
• 1: 焦虑
• 2: 中性
• 3: 积极
• 4: 讽刺

性能指标

分项准确率：

• 攻击性: 0.9120
• 焦虑: 0.9462
• 中性: 0.8663
• 积极: 0.8884
• 讽刺: 0.8426

快速调用

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

MODEL_ID = "Kostya165/rubert_tiny2_russian_emotion_sentiment"
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model     = AutoModelForSequenceClassification.from_pretrained(MODEL_ID)
model.eval()

texts = ["今天真是美好的一天！", "这一切都让我火冒三丈。"]
enc = tokenizer(texts, padding=True, truncation=True, max_length=128, return_tensors="pt")
with torch.no_grad():
    logits = model(**enc).logits
    preds = logits.argmax(dim=-1).tolist()

labels = [model.config.id2label[p] for p in preds]
print(labels)  # 示例输出: ['positive', 'aggression']

训练参数

• 基础架构: cointegrated/rubert-tiny2
• 训练数据: Kostya165/ru_emotion_dvach
• 训练轮次: 2
• 批处理量: 32
• 学习率: 1e-5
• 精度模式: FP16
• 正则化: Dropout 0.1, 权重衰减 0.01, 预热比例 0.1

运行环境

• transformers>=4.30.0
• torch>=1.10.0
• datasets
• evaluate

版权声明

CC-BY-SA 4.0.

学术引用

@article{rubert_tiny2_russian_emotion_sentiment,
  title   = {Russian Emotion Sentiment Classification with RuBERT-tiny2},
  author  = {Kostya165},
  year    = {2024},
  howpublished = {\url{https://huggingface.co/Kostya165/rubert_tiny2_russian_emotion_sentiment}}
}