Few-shot geological relationship extraction based on prompt and metric learning

doi:10.13745/j.esf.sf.2025.4.65

Abstract

Abstract:

The research in the field of geology is undergoing profound transformations, with the construction of a new knowledge system as its core and big data serving as the driving force. The construction of geological knowledge graphs can effectively address the challenge of knowledge discovery and limited reasoning in scenarios characterized by fragmented data. As one of the critical technologies for constructing knowledge graphs, relation extraction technology plays a pivotal role in identifying relationships between geological entities. Traditional relation extraction techniques are intrinsically contingent upon extensive large-scale annotated datasets. However, the intricacy and specificity of entity relationships in the geological domain render manual annotation of data laborious and time-consuming, consequently leading to a paucity of large-scale labeled datasets. Therefore, the effective implementation of traditional relation extraction techniques within the geological domain is significantly circumscribed. Given the above dilemmas, this study proposes a few-shot learning method for geological relation extraction based on the prototypical network, which innovatively introduces an enhanced prompt learning mechanism and optimizes the instance representation and relation description representation through contrastive learning, thereby significantly improving the representativeness of the prototype. Meanwhile, the weighted loss function and difficult task-assisted training strategy are adopted to enhance the model’s focus on difficult tasks, which effectively improves the overall accuracy. The experimental findings demonstrate that our approach achieves an accuracy of 82.16% in the 5-way 1-shot scenario of a geological few-shot relation extracted dataset. This represents an enhancement of 1.94% over the advanced general-domain model, SimpleFSRE, and 9.01% over the prototypical network, Proto-BERT method. These results substantiate the efficacy of our method.

Key words: few-shot learning, relation extraction, geological knowledge graph, prototype network, prompt learning

CLC Number:

P209
TP391

ZHANG Zhiting, PENG Shuai, QUE Xiang, CHEN Qiyu. Few-shot geological relationship extraction based on prompt and metric learning[J]. Earth Science Frontiers, 2025, 32(4): 250-261.

Figures/Tables 9

References 27

[1]	周成虎, 王华, 王成善, 等. 大数据时代的地学知识图谱研究[J]. 中国科学: 地球科学, 2021, 51(7): 1070-1079.
[2]	周永章, 张前龙, 黄永健, 等. 钦杭成矿带斑岩铜矿知识图谱构建及应用展望[J]. 地学前缘, 2021, 28(3): 67-75. DOI
[3]	季晓慧, 董雨航, 杨中基, 等. 基于知识图谱多跳推理的中文矿物知识问答方法与系统[J]. 地学前缘, 2024, 31(4): 37-46. DOI
[4]	黄徐胜, 朱月琴, 付立军, 等. 基于BERT的金矿地质实体关系抽取模型研究[J]. 地质力学学报, 2021, 27(3): 391-399.
[5]	邱芹军, 王斌, 徐德馨, 等. 地质领域文本实体关系联合抽取方法[J]. 高校地质学报, 2023, 29(3): 419-428.
[6]	SHIN T, RAZEGHI Y, LOGAN IV R L, et al. AutoPrompt: eliciting knowledge from language models with automatically generated prompts[C]// Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP). Online: Association for Computational Linguistics, 2020: 4222-4235.
[7]	SNELL J, SWERSKY K, ZEMEL R S. Prototypical networks for few-shot learning[J]. Advances in neural information processing systems, 2017, 30: 4080-4090.
[8]	HAN X, ZHAO W L, DING N, et al. PTR: prompt tuning with rules for text classification[J]. AI Open, 2022, 3: 182-192.
[9]	CHEN X, ZHANG N, XIE X, et al. KnowPrompt: knowledge-aware prompt-tuning with synergistic optimization for relation extraction[C]// Proceedings of the ACM Web Conference 2022. Lyon: ACM, 2022: 2778-2788.
[10]	HE K, HUANG Y C, MAO R, et al. Virtual prompt pre-training for prototype-based few-shot relation extraction[J]. Expert Systems with Applications, 2023, 213: 118927.
[11]	ZHAO X Y, YANG M, QU Q, et al. Few-shot relation extraction with autom-atically generated prompts[J]. IEEE Transactions on Neural Networks and Learning Systems, 2025, 36(3): 4971-4983.
[12]	GAO T, HAN X, LIU Z, et al. Hybrid attention-based prototypical networks for noisy few-shot relation classification[C]// Proceedings of the AAAI conference on artificial intelligence. Honolulu: AAAI Press, 2019, 33(1): 6407-6414.
[13]	YANG K, ZHENG N, DAI X, et al. Enhance prototypical network with text descriptions for few-shot relation classification[C]// Proceedings of the 29th ACM international conference on information & knowledge management. New York: ACM, 2020: 2273-2276.
[14]	HAN J, CHENG B, LU W. Exploring task difficulty for few-shot relation extraction[C]// Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Punta Cana: Association for Computational Linguistics, 2021: 2605-2616.
[15]	LIU Y, HU J, WAN X, et al. A Simple yet effective relation information guided approach for few-shot relation extraction[C]//Findings of the Association for Computational Linguistics:ACL 2022. Dublin:Association for Computational Linguistics, 2022: 757-763.
[16]	BORCHERT P, DE WEERDT J, MOENS M F. Efficient information extraction in few-shot relation classification through contrastive representation learning[C]//Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 2: Short Papers). Mexico City: Association for Computational Linguistics, 2024: 638-646.
[17]	SUN Q, LIU Y, CHUA T S, et al. Meta-transfer learning for few-shot learning[C]// Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. California: IEEE, 2019: 403-412.
[18]	LIU X, ZHENG Y, DU Z, et al. GPT understands, too[J]. AI Open, 2024, 5: 208-215.
[19]	LESTERB, AL-RFOUR, CONSTANTN. The power of scale for parameter-efficient prompt tuning[C]// Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Punta Cana: Association for Computational Linguistics, 2021: 3045-3059.
[20]	DEVLINJ, CHANG M W, LEE K, et al. BERT: pre-training of deep bidirectional transformers for language understanding[C]// Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics:Human Language Technologies. Minneapolis: Association for Computational Linguistics, 2019: 4171-4186.
[21]	孙东, 杨涛, 曹楠, 等. 泸定M_S6.8地震同震地质灾害特点及防控建议[J]. 地学前缘, 2023, 30(3): 476-493. DOI
[22]	李文佳. 面向自然语言的地质对象知识表达模型构建关键技术研究[D]. 武汉: 中国地质大学(武汉), 2022.
[23]	田苗. 地质实体关系抽取方法及应用研究[D]. 宜昌: 三峡大学, 2023.
[24]	张利军, 鲁文豪, 张建东, 等. 基于深度学习的镜下岩石、矿物薄片识别[J]. 地学前缘, 2024, 31(3): 498-510. DOI
[25]	宋轩宇, 许民, 康世昌, 等. 基于机器学习的冰冻圈典型流域水文过程模拟研究[J]. 地学前缘, 2023, 30(4): 451-469. DOI
[26]	HAN X, ZHU H, YU P, et al. FewRel: a large-scale supervised few-shot relation classification dataset with state-of-the-art evaluation[C]// Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Brussels: Association for Computational Linguistics, 2018: 4803-4809.
[27]	GAO T, HAN X, ZHU H, et al. FewRel 2.0: towards more challenging few-shot relation classification[C]// Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing. Hong Kong: Association for Computational Linguistics, 2019: 6250-6255.

关系大类	关系类别	关系描述
时空关系	时间地点演变	指某地质事件发生的具体地质年代或时间跨度。指地质事件或地质构造所在的具体位置。表示地质体或地质构造随时间的变化过程。
从属关系	属于拥有	表示某地质现象或地质单元归属于特定的构造系统。指某一地质区域或构造拥有的资源或特性。
因果关系	导致	指某一地质事件或过程引发了另一地质现象。
特征关系	地质构造属性	描述地球表面或地下的结构特征。指地质体或地质单元的固有特征。
功能关系	用途	指某一地质资源或区域的实际应用。
交互关系	响应	指地质系统对外部应力、环境变化或人类活动的反应。
认知关系	认为研究	表达地质学家基于已有的地质证据对某地质现象的解释或假设。指对地质现象、地质构造或地质资源的科学调查和分析。

关系大类	关系类别	关系描述
时空关系	时间地点演变	指某地质事件发生的具体地质年代或时间跨度。指地质事件或地质构造所在的具体位置。表示地质体或地质构造随时间的变化过程。
从属关系	属于拥有	表示某地质现象或地质单元归属于特定的构造系统。指某一地质区域或构造拥有的资源或特性。
因果关系	导致	指某一地质事件或过程引发了另一地质现象。
特征关系	地质构造属性	描述地球表面或地下的结构特征。指地质体或地质单元的固有特征。
功能关系	用途	指某一地质资源或区域的实际应用。
交互关系	响应	指地质系统对外部应力、环境变化或人类活动的反应。
认知关系	认为研究	表达地质学家基于已有的地质证据对某地质现象的解释或假设。指对地质现象、地质构造或地质资源的科学调查和分析。

	地质数据集		Fewrel
模型	5way-1shot	5way-5shot	5way-1shot	5way-5shot	10way-1shot	10way-5shot
Proto-BERT	73.15%	74.96%	84.77%	89.54%	76.85%	83.42%
BERT-PAIR	73.45%	75.02%	85.66%	89.48%	76.84%	81.76%
HCRP	79.17%	80.28%	90.90%	93.22%	84.11%	87.79%
SimpleFSRE	80.22%	81.14%	91.29%	94.05%	86.09%	89.68%
GFSRE	82.16%	82.88%	92.98%	94.39%	86.71%	90.23%

	地质数据集		Fewrel
模型	5way-1shot	5way-5shot	5way-1shot	5way-5shot	10way-1shot	10way-5shot
Proto-BERT	73.15%	74.96%	84.77%	89.54%	76.85%	83.42%
BERT-PAIR	73.45%	75.02%	85.66%	89.48%	76.84%	81.76%
HCRP	79.17%	80.28%	90.90%	93.22%	84.11%	87.79%
SimpleFSRE	80.22%	81.14%	91.29%	94.05%	86.09%	89.68%
GFSRE	82.16%	82.88%	92.98%	94.39%	86.71%	90.23%

Model	5way-5shot
GFSRE	82.88%
去除增强提示学习模板	81.04%
去除对比学习模块	80.48%
去除困难任务的加权损失与辅助训练任务	81.38%