-
Numerous data and knowledge generated and stored as text in peer-reviewed scientific literature are important for materials research and development. Although text mining can automatically explore this information, the barriers of acquiring high-quality textual data prevent its general application in materials science. Herein, we systematically analyze the issues of textual DATA QUALITY and related research from the perspectives of data quality and quantity. Following this, we propose a pipeline to construct high-quality datasets for text mining in materials science. In this pipeline, we utilize the traceable automatic acquisition scheme of literature to ensure the traceability of textual data. Then, a data processing method driven by downstream tasks is used to generate high-quality pre-annotated corpora conditioned on the characteristics of material texts. On this basis, we define a general annotation scheme derived from materials science tetrahedron to complete high-quality annotation. Finally, a conditional data augmentation model incorporating material domain knowledge (cDA-DK) is constructed to augment the data quantity. Experimental results on datasets with various material systems demonstrate that our method can effectively improve the accuracy of downstream models and the F1-score towards the named entity recognition task in NASICON-type solid electrolyte material reaches 84%. This study provides an important insight into the general application of text mining in materials science, and is expected to advance the material design and discovery driven by data and knowledge bidirectionally.
-
Keywords:
- text mining in materials science /
- data augmentation /
- data quality
[1] Gupta T, Zaki M, Krishnan N M A, Mausam 2022 npj Comput. Mater. 8 102
Google Scholar
[2] Olivetti E A, Cole J M, Kim E, Kononova O, Ceder G, Han T Y J, Hiszpanski A M 2020 Appl. Phys. Rev. 7 041317
Google Scholar
[3] Venugopal V, Sahoo S, Zaki M, Agarwal M, Gosvami N N, Krishnan N M A 2021 Patterns 2 100290
Google Scholar
[4] Kononova O, He T, Huo H, Trewartha A, Olivetti E A, Ceder G 2021 iScience 24 102155
Google Scholar
[5] Kim E, Huang K, Saunders A, McCallum A, Ceder G, Olivetti E 2017 Chem. Mater. 29 9436
Google Scholar
[6] Mysore S, Jensen Z, Kim E, Huang K, Chang H S, Strubell E, Flanigan J, McCallum A, Olivetti E 2019 Proceedings of the 13th Linguistic Annotation Workshop Florence, Italy, August 1, 2019 p56
[7] Tshitoyan V, Dagdelen J, Weston L, Dunn A, Rong Z, Kononova O, Persson K A, Ceder G, Jain A 2019 Nature 571 95
Google Scholar
[8] Vaucher A C, Zipoli F, Geluykens J, Nair V H, Schwaller P, Laino T 2020 Nat. Commun. 11 3601
Google Scholar
[9] Nie Z, Zheng S, Liu Y, Chen Z, Li S, Lei K, Pan F 2022 Adv. Funct. Mater. 32 2201437
Google Scholar
[10] Wang W R, Jiang X, Tian S H, Liu P, Dang D P, Su Y J, Lookman T, Xie J X 2022 npj Comput. Mater. 8 9
Google Scholar
[11] Weston L, Tshitoyan V, Dagdelen J, Kononova O, Trewartha A, Persson K A, Ceder G, Jain A 2019 J. Chem. Inf. Model. 59 3692
Google Scholar
[12] Friedrich A, Adel H, Tomazic F, Hingerl J, Benteau R, Maruscyk A, Lange L 2020 Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics Seattle, Washington, July 5–10, 2020 p1255
[13] He T, Sun W, Huo H, Kononova O, Rong Z, Tshitoyan V, Botari T, Ceder G 2020 Chem. Mater. 32 7861
Google Scholar
[14] Beal M S, Hayden B E, Le Gall T, Lee C E, Lu X, Mirsaneh M, Mormiche C, Pasero D, Smith D C, Weld A, Yada C, Yokoishi S 2011 ACS Comb. Sci. 13 375
Google Scholar
[15] Rajan A C, Mishra A, Satsangi S, Vaish R, Mizuseki H, Lee K R, Singh A K 2018 Chem. Mater. 30 4031
Google Scholar
[16] 刘悦, 邹欣欣, 杨正伟, 施思齐 2022 硅酸盐学报 50 863
Google Scholar
Liu Y, Zou X X, Yang Z W, Shi S Q 2022 J. Chin. Ceram. Soc. 50 863
Google Scholar
[17] 赵凯琳, 靳小龙, 王元卓 2021 软件学报 32 349
Google Scholar
Zhao K L, Jin X L, Wang Y Z 2021 J. Software 32 349
Google Scholar
[18] Wei J, Zou K 2019 Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9 th International Joint Conference on Natural Language Processing Hong Kong, China, November 3–7, 2019 p6382
[19] Morris J X, Lifland E, Yoo J Y, Grigsby J, Jin D, Qi Y 2020 Proceedings of the 2020 EMNLP (Systems Demonstrations) Punta Cana, Dominican Republic, November 16–20, 2020 p119
[20] Malandrakis N, Shen M, Goyal A, Gao S, Sethi A, Metallinou A 2019 Proceedings of the 3rd Workshop on Neural Gene ration and Translation (WNGT 2019) Hong Kong, China, November 4, 2019 p90
[21] Wu X, Lü S W, Zang L J, Han J Z, Hu S L 2019 Computational Science–ICCS 2019 (Cham: Springer Nature Switzerland AG) p84
[22] Kumar V, Choudhary A, Cho E 2021 arXiv: 2003.02245 [cs. CL]
[23] Xu X, Lei Y, Li Z 2020 IEEE Trans. Ind. Electron. 67 2326
Google Scholar
[24] Shinyama Yhttps://euske.github.io/pdfminer/ [2022-11-20]
[25] Jessop D M, Adams S E, Willighagen E L, Hawizy L, Murray-Rust P 2011 J. Cheminf. 3 41
Google Scholar
[26] Hawizy L, Jessop D M, Adams N, Murray-Rust P 2011 J. Cheminf. 3 17
Google Scholar
[27] Swain M C, Cole J M 2016 J. Chem. Inf. Model. 56 1894
Google Scholar
[28] Sun C C 2009 J. Pharm. Sci. 98 1671
Google Scholar
[29] Armstrong S, Church K, Isabelle P, Manzi S, Tzoukermann E, Yarowsky D 1999 Natural Language Processing Using Very Large Corpora (Berlin: Springer) pp157–176
[30] Liu Y, Ott M, Goyal N, Du J, Joshi M, Chen D, Levy O, Lewis M, Zettlemoyer L, Stoyanov V 2019 arXiv: 1907.11692 [cs. CL]
[31] Chen S, Wu C, Shen L, Zhu C, Huang Y, Xi K, Maier J, Yu Y 2017 Adv. Mater. 29 1700431
Google Scholar
[32] 肖睿娟, 李泓, 陈立泉 2018 物理学报 67 128801
Google Scholar
Xiao R J, Li H, Chen L Q 2018 Acta Phys. Sin. 67 128801
Google Scholar
[33] Liu Y, Ge X Y, Yang Z W, Sun S Y, Liu D H, Avdeev M, Shi S Q 2022 J. Power Sources 545 231946
Google Scholar
-
图 1 高质量材料文本挖掘数据集构建管道
Figure 1. The pipeline for constructing high-quality datasets for materials text mining.
图 2 文献的数据与过程溯源示意图
Figure 2. The illustration of the traceability of literature data and process.
图 3 实体关系标注流程示意图
Figure 3. The process of annotation on entities and relations.
图 4 基于cDA-DK的材料文本数据增强
Figure 4. Materials textual data augmentation based on cDA-DK.
图 5 两份数据集的样本统计情况对比 (a) 三元组个数分布情况; (b) 语句长度分布情况
Figure 5. Comparison of sample statistics of two datasets: (a) The distribution of numbers of triplets; (b) the distribution of length of sentence.
图 6 实体识别模型在不同数据集上的混淆矩阵 (a) Dataset 1的混淆矩阵; (b) Dataset 2的混淆矩阵
Figure 6. Confusion matrix of NER model on various datasets: (a) The confusion matrix of Dataset 1; (b) the confusion matrix of Dataset 2.
图 7 MatBERT-BiLSTM-CRF在不同数据集上的训练及验证Loss变化曲线 (a) Dataset 1上的Loss变化曲线; (b) Dataset 2上的Loss变化曲线; (c) Dataset 4上的Loss变化曲线; (d) Dataset 5上的Loss变化曲线
Figure 7. The training and validation loss function of MatBERT-BiLSTM-CRF on various datasets: (a) The loss function on Dataset 1; (b) the loss function on Dataset 2; (c) the loss function on Dataset 4; (d) the loss function on Dataset 5.
表 1 材料科学文本语料获取方式对比
Table 1. Comparison of acquisition methods of materials scientific corpus.
获取方式 数据库 文档类型 访问权限 文档数量 参考 索引数据库 API CAplus 论文, 专利, 报告 订阅 少 www.cas.org/support/documentation/references DOAJ 论文 部分订阅 少 doaj.org PubMed Central 论文 开放获取 较少 www.ncbi.nlm.nih.gov/pmc Science Direct 论文 订阅 少 dev.elsevier.com/api_docs.html Scopus 摘要 开放获取 较少 Springer Nature 论文, 书籍 订阅 少 dev.springernature.com/ 网络爬虫 网页 论文, 专利, 报告, 书籍 开放获取 多 requests.readthedocs.io, crummy.com/software/BeautifulSoup 
DownLoad: CSV
表 2 化学与材料科学中常用的自然语言处理工具
Table 2. Common natural language processing tools in chemistry and materials science.
DownLoad: CSV
表 3 已有材料文本挖掘研究中的实体标签定义对比
Table 3. Comparison of entity label definitions in previous materials text mining research.
来源 目标 标签数 标签类别 适用领域 应用实例 Weston等[11] 构建材料领域最新研究结果
与历史文献的关联7 无机材料, 相结构, 描述符, 属
性, 应用, 合成方法, 表征方法无机材料 目标材料检索, 文献搜索
与总结, 元信息分析He等[13] 从无机固相合成反应文献
中挖掘反应前体信息3 材料, 合成反应前体,
目标化合物无机固相
合成反应固相合成反应前体
数据挖掘, 元信息分析Friedrich等[12] 标注科学出版物中与SOFCs
实验相关的信息4(SOFC) 17(SOFC-slot) 实验, 材料, 数值, 应用等 电池材料 构建SOFCs科学语料库并用
于多个实验信息提取任务Wang等[10] 从文献中自动挖掘出数据驱动的材
料设计模型所需的高质量可靠数据6 元素, 合金命名实体, 成分含
量, 属性描述符, 属性值, 其他合金材料 钴基单晶高温合金${ {\rm{\gamma } } }'$
相固溶温度预测Nie等[9] 构建语义表示框架以探索潜在
的锂离子电池阴极材料3 无机材料, 锂离子电池
阴极材料, 属性描述符电池材料 新型锂离子电池阴
极材料设计与寻优
DownLoad: CSV
表 4 面向通用领域的材料实体类型定义
Table 4. The definition of materials entity types in the general domain.
实体标签 定义 示例 Composition 与化学式有关的内容; 描述材料内部与含量相关的内容等. NaCl, CaCl2; Na concentration, Electrons charge carriers. Structure 晶体结构; 相; 用于刻画晶体结构的名称等. Fcc, Phase; Bottleneck, Channel, Path. Property 带单位的可度量值; 材料表现出来定性的性质或现象;
描述材料产生物理/化学行为或物理/化学机制的名词等.Conductivity, Activation, Radius; Ferroelectric, Metallic; Phase transition, Ionic reaction. Processing 材料合成技术或加工工艺; 材料改性手段等. Solid state reaction, Annealing; Doping. Characterization 用于表征材料的任何实验、理论、模型或公式等. XRD, STM, Photoluminescence, DFT;
Bethe-Salpeter equation.Application 任何高级的应用; 任何特定的器件、系统等. Cathode, Photovoltaics; Battery Management System. Feature 样品类型、形状的特殊描述. Single crystal, Bulk, nanotube, Quantum dot. Condition 描述材料所处的环境或外部条件. 980 $°{\rm{C}}$, 1000 MPa.
DownLoad: CSV
表 5 面向通用领域的材料实体关系类型定义
Table 5. The definition of materials relation types in the general domain.
关系标签 (A to B) 定义 可能存在此关系的实体类型 Cause-Effect A对B有影响 Property-Property, Composition-Structure, Structure-Property, ... Component-Whole A是B的部分 Composition-Composition, ... Feature-Of A是B的特征 Feature-Composition, Feature-Application, ... Located-Of A占据了B位置 Composition-Structure, ... Instance-Of A是B的实例 Composition-Composition, Structure-Structure, Property-Property, ... Condition-On A的条件是B Processing-Condition, ... Method-Of A的表征方法是B Property-Characterization, ... Other A与B存在除上述关系类型外的其他关系 —
DownLoad: CSV
表 6 常用文本标注工具对比
Table 6. Comparison of common tools for text annotation.
标注工具 适配任务 文本要求 角色管理权限 难易性 友好性 可扩展性 参考 Label Studio 多模态信息标注 严格 不完善 普通 中 中 labelstud.io Brat 关系标注 一般 完善 普通 中 低 github.com/nlplab/brat Doccano 文本分类 严格 较完善 普通 低 低 github.com/doccano EasyData 实体与关系标注 一般 完善 容易 高 高 ai.baidu.com/easydata/
DownLoad: CSV
算法1 数据增强方法cDA-DK 输入 原始数据集$ {D}_{{\rm{t}}{\rm{r}}{\rm{a}}{\rm{i}}{\rm{n}}} = \{\left({x}_{1}, {y}_{1}\right), \left({x}_{2}, {y}_{2}\right), \dots , \left({x}_{n}, {y}_{n}\right)\} $ 预训练语言模型模型$ {P}_{{\rm{D}}{\rm{i}}{\rm{s}}{\rm{t}}{\rm{i}}{\rm{l}}{\rm{R}}{\rm{o}}{\rm{B}}{\rm{E}}{\rm{R}}{\rm{T}}{\rm{a}}} $ 材料领域词典$ C=\{{w}_{1}, {w}_{2}, \dots , {w}_{m}\} $输出 增强数据集$ {D}_{{\rm{s}}{\rm{y}}{\rm{n}}{\rm{t}}{\rm{h}}{\rm{e}}{\rm{t}}{\rm{i}}{\rm{c}}} $ 1: 开始 2: for $ {w}_{i}\in C $ do3: $ {w}_{i} $输入至$ {P}_{{\rm{D}}{\rm{i}}{\rm{s}}{\rm{t}}{\rm{i}}{\rm{l}}{\rm{R}}{\rm{o}}{\rm{B}}{\rm{E}}{\rm{R}}{\rm{T}}{\rm{a}}} $的词汇表并训练其对应的词向量 4: 在下游任务文本数据增强上微调$ {P}_{{\rm{D}}{\rm{i}}{\rm{s}}{\rm{t}}{\rm{i}}{\rm{l}}{\rm{R}}{\rm{o}}{\rm{B}}{\rm{E}}{\rm{R}}{\rm{T}}{\rm{a}}} $得到$ {F}_{{\rm{D}}{\rm{i}}{\rm{s}}{\rm{t}}{\rm{i}}{\rm{l}}{\rm{R}}{\rm{o}}{\rm{B}}{\rm{E}}{\rm{R}}{\rm{T}}{\rm{a}}} $ 5: 初始化$ {D}_{{\rm{s}}{\rm{y}}{\rm{n}}{\rm{t}}{\rm{h}}{\rm{e}}{\rm{t}}{\rm{i}}{\rm{c}}}=\left\{\right\} $ 6: for $ \left\{{x}_{i}, {y}_{i}\right\}\in {D}_{{\rm{t}}{\rm{r}}{\rm{a}}{\rm{i}}{\rm{n}}} $ do 7: $ ({\widehat{x}}_{i}, {\widehat{y}}_{i})={F}_{{\rm{D}}{\rm{i}}{\rm{s}}{\rm{t}}{\rm{i}}{\rm{l}}{\rm{R}}{\rm{o}}{\rm{B}}{\rm{E}}{\rm{R}}{\rm{T}}{\rm{a}}}({x}_{i}, {y}_{i}) $ // 生成新的样本 8: $ {D}_{{\rm{s}}{\rm{y}}{\rm{n}}{\rm{t}}{\rm{h}}{\rm{e}}{\rm{t}}{\rm{i}}{\rm{c}}}={D}_{{\rm{s}}{\rm{y}}{\rm{n}}{\rm{t}}{\rm{h}}{\rm{e}}{\rm{t}}{\rm{i}}{\rm{c}}}\cup ({\widehat{x}}_{i}, {\widehat{y}}_{i}) $ // 生成样本加入增强数据集 9: 结束
DownLoad: CSV
表 7 NASICON实体关系数据集与CoNLL-2004数据集的对比
Table 7. Comparison of the NASICON dataset with the CoNLL-2004 dataset.
数据集 样本数 实体类型 实体数 关系类型 关系数 CoNLL-2004 1, 441 4 5, 347 5 2, 020 NASICON 2, 434 8 4, 857 8 2, 297
DownLoad: CSV
表 8 NASICON实体关系数据集在增强前后的数据示例对比
Table 8. Comparison of samples before and after augmentation of NASICON dataset.
数据集 样本数 实体数 关系数 示例 原始数据集 2434 4857 2297 The (O) ionic (B-Property) conductivity (I-Property) decreases (O) with (O) increasing (O) activation (B-Property) energy (I-Property) . (O) cDA-DK 增强数据集 4846 9714 4594 The (O) electrode (B-Property) conductivity (I-Property) decreases (O) with (O) increasing (O) electric (B-Property) energy (I-Property) . (O)
DownLoad: CSV
表 9 实验数据集信息
Table 9. The details of experimental datasets.
数据集名称 应用领域 重命名 样本量 语料规模 来源 NASICON 实体识别数据集 NASICON 型固态电解质 Dataset 1 2, 434 55篇文献 领域专家标注 Dataset 2 2, 434 — 数据增强 Dataset 3 305 35篇文献 非专业人员标注 Matscholar[11] 无机材料 Dataset 4 5, 459 800份摘要 领域专家标注 Dataset 5 5, 459 — 数据增强
DownLoad: CSV
表 10 实体识别模型在不同材料数据集上的实验结果
Table 10. The results of NER model on various materials datasets.
数据集 材料类别 样本量 Precision Recall F1-score Dataset 1 NASICON 型固态电解质 2, 434 0.78 0.83 0.80 Dataset 2 2, 434 0.68 0.72 0.70 Dataset 2+3 2, 739 0.83 0.85 0.84 Dataset 4 无机材料 5, 459 0.86 0.90 0.88 Dataset 5 5, 459 0.75 0.78 0.77
DownLoad: CSV
-
[1] Gupta T, Zaki M, Krishnan N M A, Mausam 2022 npj Comput. Mater. 8 102
Google Scholar
[2] Olivetti E A, Cole J M, Kim E, Kononova O, Ceder G, Han T Y J, Hiszpanski A M 2020 Appl. Phys. Rev. 7 041317
Google Scholar
[3] Venugopal V, Sahoo S, Zaki M, Agarwal M, Gosvami N N, Krishnan N M A 2021 Patterns 2 100290
Google Scholar
[4] Kononova O, He T, Huo H, Trewartha A, Olivetti E A, Ceder G 2021 iScience 24 102155
Google Scholar
[5] Kim E, Huang K, Saunders A, McCallum A, Ceder G, Olivetti E 2017 Chem. Mater. 29 9436
Google Scholar
[6] Mysore S, Jensen Z, Kim E, Huang K, Chang H S, Strubell E, Flanigan J, McCallum A, Olivetti E 2019 Proceedings of the 13th Linguistic Annotation Workshop Florence, Italy, August 1, 2019 p56
[7] Tshitoyan V, Dagdelen J, Weston L, Dunn A, Rong Z, Kononova O, Persson K A, Ceder G, Jain A 2019 Nature 571 95
Google Scholar
[8] Vaucher A C, Zipoli F, Geluykens J, Nair V H, Schwaller P, Laino T 2020 Nat. Commun. 11 3601
Google Scholar
[9] Nie Z, Zheng S, Liu Y, Chen Z, Li S, Lei K, Pan F 2022 Adv. Funct. Mater. 32 2201437
Google Scholar
[10] Wang W R, Jiang X, Tian S H, Liu P, Dang D P, Su Y J, Lookman T, Xie J X 2022 npj Comput. Mater. 8 9
Google Scholar
[11] Weston L, Tshitoyan V, Dagdelen J, Kononova O, Trewartha A, Persson K A, Ceder G, Jain A 2019 J. Chem. Inf. Model. 59 3692
Google Scholar
[12] Friedrich A, Adel H, Tomazic F, Hingerl J, Benteau R, Maruscyk A, Lange L 2020 Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics Seattle, Washington, July 5–10, 2020 p1255
[13] He T, Sun W, Huo H, Kononova O, Rong Z, Tshitoyan V, Botari T, Ceder G 2020 Chem. Mater. 32 7861
Google Scholar
[14] Beal M S, Hayden B E, Le Gall T, Lee C E, Lu X, Mirsaneh M, Mormiche C, Pasero D, Smith D C, Weld A, Yada C, Yokoishi S 2011 ACS Comb. Sci. 13 375
Google Scholar
[15] Rajan A C, Mishra A, Satsangi S, Vaish R, Mizuseki H, Lee K R, Singh A K 2018 Chem. Mater. 30 4031
Google Scholar
[16] 刘悦, 邹欣欣, 杨正伟, 施思齐 2022 硅酸盐学报 50 863
Google Scholar
Liu Y, Zou X X, Yang Z W, Shi S Q 2022 J. Chin. Ceram. Soc. 50 863
Google Scholar
[17] 赵凯琳, 靳小龙, 王元卓 2021 软件学报 32 349
Google Scholar
Zhao K L, Jin X L, Wang Y Z 2021 J. Software 32 349
Google Scholar
[18] Wei J, Zou K 2019 Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9 th International Joint Conference on Natural Language Processing Hong Kong, China, November 3–7, 2019 p6382
[19] Morris J X, Lifland E, Yoo J Y, Grigsby J, Jin D, Qi Y 2020 Proceedings of the 2020 EMNLP (Systems Demonstrations) Punta Cana, Dominican Republic, November 16–20, 2020 p119
[20] Malandrakis N, Shen M, Goyal A, Gao S, Sethi A, Metallinou A 2019 Proceedings of the 3rd Workshop on Neural Gene ration and Translation (WNGT 2019) Hong Kong, China, November 4, 2019 p90
[21] Wu X, Lü S W, Zang L J, Han J Z, Hu S L 2019 Computational Science–ICCS 2019 (Cham: Springer Nature Switzerland AG) p84
[22] Kumar V, Choudhary A, Cho E 2021 arXiv: 2003.02245 [cs. CL]
[23] Xu X, Lei Y, Li Z 2020 IEEE Trans. Ind. Electron. 67 2326
Google Scholar
[24] Shinyama Yhttps://euske.github.io/pdfminer/ [2022-11-20]
[25] Jessop D M, Adams S E, Willighagen E L, Hawizy L, Murray-Rust P 2011 J. Cheminf. 3 41
Google Scholar
[26] Hawizy L, Jessop D M, Adams N, Murray-Rust P 2011 J. Cheminf. 3 17
Google Scholar
[27] Swain M C, Cole J M 2016 J. Chem. Inf. Model. 56 1894
Google Scholar
[28] Sun C C 2009 J. Pharm. Sci. 98 1671
Google Scholar
[29] Armstrong S, Church K, Isabelle P, Manzi S, Tzoukermann E, Yarowsky D 1999 Natural Language Processing Using Very Large Corpora (Berlin: Springer) pp157–176
[30] Liu Y, Ott M, Goyal N, Du J, Joshi M, Chen D, Levy O, Lewis M, Zettlemoyer L, Stoyanov V 2019 arXiv: 1907.11692 [cs. CL]
[31] Chen S, Wu C, Shen L, Zhu C, Huang Y, Xi K, Maier J, Yu Y 2017 Adv. Mater. 29 1700431
Google Scholar
[32] 肖睿娟, 李泓, 陈立泉 2018 物理学报 67 128801
Google Scholar
Xiao R J, Li H, Chen L Q 2018 Acta Phys. Sin. 67 128801
Google Scholar
[33] Liu Y, Ge X Y, Yang Z W, Sun S Y, Liu D H, Avdeev M, Shi S Q 2022 J. Power Sources 545 231946
Google Scholar
-
补充材料-7-20222316-070701.pdf
DownLoad:
Catalog
Metrics
- Abstract views: 6503
- PDF Downloads: 217
- Cited By: 0
