Data imbalance presents a significant challenge in various machine learning (ML) tasks, particularly named entity recognition (NER) within natural language processing (NLP). NER exhibits a data imbalance with a long-tail distribution, featuring numerous minority classes (i.e., entity classes) and a single majority class (i.e., O-class). The imbalance leads to the misclassifications of the entity classes as the O-class. To tackle the imbalance, we propose a simple and effective learning method, named majority or minority (MoM) learning. MoM learning incorporates the loss computed only for samples whose ground truth is the majority class (i.e., the O-class) into the loss of the conventional ML model. Evaluation experiments on four NER datasets (Japanese and English) showed that MoM learning improves prediction performance of the minority classes, without sacrificing the performance of the majority class and is more effective than widely known and state-of-the-art methods. We also evaluated MoM learning using frameworks as sequential labeling and machine reading comprehension, which are commonly used in NER. Furthermore, MoM learning has achieved consistent performance improvements regardless of language, model, or framework.

翻译：数据不平衡是机器学习（ML）任务中的重大挑战，尤其在自然语言处理（NLP）的命名实体识别（NER）领域。NER存在长尾分布的数据不平衡现象，表现为大量少数类（即实体类别）与单一多数类（即O类）并存。这种不平衡会导致实体类别被误判为O类。为解决该问题，我们提出一种简单有效的学习方法——多数或少数（MoM）学习。MoM学习将仅针对真实标签为多数类（即O类）的样本计算的损失，融入传统ML模型的损失函数中。在四个NER数据集（日语和英语）上的评估实验表明，MoM学习能在不牺牲多数类性能的前提下提升少数类的预测效果，且优于广泛使用的及当前最先进的方法。我们还采用NER中常用的序列标注与机器阅读理解框架对MoM学习进行了评估。此外，MoM学习在语言、模型或框架层面均能实现一致的性能提升。