对比三类模型在细粒度情绪识别上的表现，发现不同模型各有优势。
Fine-Grained Emotion Detection on GoEmotions: Experimental Comparison of Classical Machine Learning, BiLSTM, and Transformer Models
- 用逻辑回归、BiLSTM和BERT分别处理情绪分类任务。
- BERT在罕见情绪识别上表现最佳，宏平均F1达0.49。
- 表面词汇对常见情绪更有效，上下文信息提升稀有情绪识别。
细粒度情绪识别因标签重叠和类别不平衡而具有挑战性。本文在GoEmotions数据集上对比三类模型：基于TF-IDF的逻辑回归（二元相关训练）、带注意力的BiLSTM，以及微调的BERT模型。实验采用官方划分的训练/验证/测试集，并通过逆频率类别权重缓解不平衡问题。在Micro-F1、Macro-F1、Hamming Loss和子集准确率等多个指标上，逻辑回归取得最高Micro-F1（0.51），而BERT表现最均衡，超越原论文结果：Macro-F1达0.49，Hamming Loss为0.036，子集准确率为0.36。结果表明，常见情绪多依赖表面词汇线索，而上下文表示有助于提升稀有及模糊样本的识别性能。
原文摘要 · Abstract (English)
Fine-grained emotion recognition is a challenging multi-label NLP task due to label overlap and class imbalance. In this work, we benchmark three modeling families on the GoEmotions dataset: a TF-IDF-based logistic regression system trained with binary relevance, a BiLSTM with attention, and a BERT model fine-tuned for multi-label classification. Experiments follow the official train/validation/test split, and imbalance is mitigated using inverse-frequency class weights. Across several metrics, namely Micro-F1, Macro-F1, Hamming Loss, and Subset Accuracy, we observe that logistic regression attains the highest Micro-F1 of 0.51, while BERT achieves the best overall balance surpassing the official paper's reported results, reaching Macro-F1 0.49, Hamming Loss 0.036, and Subset Accuracy 0.36. This suggests that frequent emotions often rely on surface lexical cues, whereas contextual representations improve performance on rarer emotions and more ambiguous examples.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。