文章背景与核心概要

科学命名实体识别（SciNER）是从学术文本中提取知识的核心环节。尽管大语言模型（LLMs）在此类任务中展现出巨大潜力，但当提示词中包含大量且复杂的候选实体类型时，它们往往难以应对。为解决这一难题，作者提出了 TdSciNER，这是一个采用实体类型过滤器、多任务学习辅助分类任务以及新颖示范选择策略的类型驱动框架。

该方法通过精确过滤候选类型并激活大模型的上下文学习能力，在多个数据集上实现了媲美全监督模型的性能。本文的研究成果对提升垂直领域（如科学文献）的信息抽取准确率具有重要意义，同时也为优化大模型在复杂标签空间下的提示词工程提供了新的思路。

Enhancing Scientific Named Entity Recognition via Large Language Models: A Type-driven Multi-task Learning Approach

Enhancing Scientific Named Entity Recognition via Large Language Models: A Type-driven Multi-task Learning Approach

Summary

Summary

科学命名实体识别（SciNER）对于从学术文本中提取知识至关重要。尽管大语言模型（LLMs）展现出了广阔的应用前景，但当提示词中提供过多且复杂的候选实体类型时，它们往往难以准确处理。为了克服这一局限，作者提出了 TdSciNER，这是一种类型驱动的框架，它利用实体类型过滤器、多任务学习辅助分类任务以及新颖的示范选择策略。该方法在多个数据集上取得了可与全监督模型相媲美的性能。

Scientific named entity recognition (SciNER) is vital for extracting knowledge from academic texts. While large language models (LLMs) show promise, they often struggle when provided with an overwhelming number of complex candidate entity types in prompts. To overcome this, the authors propose TdSciNER, a type-driven framework that utilizes an entity type filter, a multi-task learning auxiliary classification task, and a novel demonstration selection strategy. This approach achieves performance comparable to fully supervised models across multiple datasets.

Article Metadata

Article Metadata

Field	Details
Authors	Tong Bao, Yi Zhao, Heng Zhang, Chengzhi Zhang
Subjects	Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Digital Libraries (cs.DL); Information Retrieval (cs.IR)
arXiv ID	arXiv:2608.08636
DOI	10.48550/arXiv.2608.08636
Journal Reference	Expert Systems With Applications, 2026
Submission History	[v1] Sun, 9 Aug 2026 [v2] Fri, 14 Aug 2026 (Current Version)

Abstract

Abstract

科学命名实体识别（SciNER）在科学文本的信息抽取和知识发现中发挥着至关重要的作用。近期，大语言模型（LLMs）已经展示出仅需少量人工干预即可实现具有竞争力的 SciNER 性能的能力。现有研究突显了结合候选实体类型信息对于大模型进行准确实体识别和分类的重要性。

Scientific named entity recognition (SciNER) plays a crucial role in information extraction and knowledge discovery from scientific texts. Recently, large language models (LLMs) have demonstrated the capacity to achieve competitive SciNER performance with minimal human effort. Existing research highlights the importance of incorporating candidate entity type information for accurate entity recognition and classification by LLMs.

然而，当提示词中提供过多的候选实体类型时，大模型在识别和标注科学文本中的实体时就会遇到困难，因为科学领域的实体类型比通用领域更为复杂。为了应对这一挑战，我们提出了 TdSciNER，这是一种类型驱动的方法，能够有效利用实体类型信息来增强 SciNER 的性能。

However, when too many candidate entity types are provided in the prompt, LLMs struggle to accurately recognize and label entities in scientific texts, where entity types are more complex than in general domains. To address this challenge, we propose TdSciNER, a type-driven approach that effectively leverages entity type information to enhance SciNER performance.

TdSciNER 的核心组件：

Key Components of TdSciNER:

1. 实体类型过滤模型： 识别给定句子中最可能存在的实体类型。
2. 辅助多类实体类型化任务： 在多任务学习框架中与 SciNER 相集成，以获得更丰富的上下文表示。
3. 新颖的示范选择策略： 基于句子相似度和实体类型多样性，以激活大语言模型的上下文学习能力，从而提高跨多个科学领域的识别准确率。

1. Entity Type Filter Model: Identifies the most likely entity types present in a given sentence.
2. Auxiliary Multi-class Entity Typing Task: Integrated within a multi-task learning framework alongside SciNER to obtain richer contextual representations.
3. Novel Demonstration Selection Strategy: Based on sentence similarity and entity type diversity to activate the in-context learning capabilities of LLMs, improving recognition accuracy across diverse scientific domains.

在三个数据集上的实验表明，我们的方法达到了与全监督模型相当的性能，这验证了 TdSciNER 的每个组件都对 SciNER 性能的提升作出了实质性贡献。

Experiments on three datasets demonstrate that our method achieves performance comparable to fully supervised models, validating that each component of TdSciNER contributes meaningfully to SciNER performance improvements.

Access & Resources

Access & Resources

• 全文： 查看 PDF
• 许可证： 知识共享署名 4.0
• 外部链接：
• Google Scholar
• Semantic Scholar
• NASA ADS

• Full-Text: View PDF
• License: Creative Commons Attribution 4.0
• External Links:
• Google Scholar
• Semantic Scholar
• NASA ADS