Tired of Sifting Through Science Papers? This AI Knowledge Graph Does It for You

Authors: (1) Yanpeng Ye, School of Computer Science and Engineering, University of New South Wales, Kensington, NSW, Australia, GreenDynamics Pty. Ltd, Kensington, NSW, Australia, and these authors contributed equally to this work; (2) Jie Ren, GreenDynamics Pty. Ltd, Kensington, NSW, Australia, Department of Materials Science and Engineering, City University of Hong Kong, Hong Kong, China, and these authors contributed equally to this work; (3) Shaozhou Wang, GreenDynamics Pty. Ltd, Kensington, NSW, Australia (shaozhou@greendynamics.com.au); (4) Yuwei Wan, GreenDynamics Pty. Ltd, Kensington, NSW, Australia and Department of Linguistics and Translation, City University of Hong Kong, Hong Kong, China; (5) Imran Razzak, School of Computer Science and Engineering, University of New South Wales, Kensington, NSW, Australia; (6) Tong Xie, GreenDynamics Pty. Ltd, Kensington, NSW, Australia and School of Photovoltaic and Renewable Energy Engineering, University of New South Wales, Kensington, NSW, Australia (tong@greendynamics.com.au); (7) Wenjie Zhang, School of Computer Science and Engineering, University of New South Wales, Kensington, NSW, Australia (wenjie.zhang@unsw.edu.au). Editor’s note: This article is part of a broader study. You’re reading Part 1 of 9. Read the rest below. Table of Links Abstract and Introduction Methods Data preparation and schema design LLMs training, evaluation and inference Entity resolution Knowledge graph construction Result Discussion Conclusion and References ABSTRACT The convergence of materials science and artificial intelligence has unlocked new opportunities for gathering, analyzing, and generating novel materials sourced from extensive scientific literature. Despite the potential benefits, persistent challenges such as manual annotation, precise extraction, and traceability issues remain. Large language models have emerged as promising solutions to address these obstacles. This paper introduces Functional Materials Knowledge Graph (FMKG), a multidisciplinary materials science knowledge graph. Through the utilization of advanced natural language processing techniques, extracting millions of entities to form triples from a corpus comprising all high-quality research papers published in the last decade. It organizes unstructured information into nine distinct labels, covering Name, Formula, Acronym, Structure/Phase, Properties, Descriptor, Synthesis, Characterization Method, Application, and Domain, seamlessly integrating papers’ Digital Object Identifiers. As the latest structured database for functional materials, FMKG acts as a powerful catalyst for expediting the development of functional materials and a fundation for building a more comprehensive material knowledge graph using full paper text. Furthermore, our research lays the groundwork for practical text-mining-based knowledge management systems, not only in intricate materials systems but also applicable to other specialized domains. Introduction In the contemporary information era, despite notable advancements, the creation and advancement of novel materials still heavily rely on traditional, time-consuming trial-and-error methods intertwined with chemical and physical intuitions. These conventional research approaches significantly impede the life-cycle of high-performance material research. Given the specialization, inherent complexity, and vast knowledge base of material science, researchers focusing on a single direction often struggle to efficiently access and understand material knowledge from multidisciplinary studies. For instance, researchers in solar cell development might not fully comprehend studies related to solid-state batteries or organic light-emitting diodes. Yet, the electronic properties of materials across these different domains are highly related, and researchers in different domains can potentially learn from each other. To accelerate the progress of materials research, there is a pressing need to efficiently integrate knowledge from various disciplines[1]. However, this vital knowledge is scattered across a vast array of over 10 million scientific papers, covering diverse topics and disciplines such as materials preparation and functionalization methods, advanced materials characterization techniques, and the exploration of physical, chemical, and biological properties, along with their applications in fields like electronic devices, clean energy storage and transfer, and mechanical engineering. This fragmentation of knowledge represents a significant barrier to interdisciplinary collaboration and innovation. A critical gap in current research infrastructure is the lack of an effective materials science database that can consolidate this scattered knowledge, facilitating easier access and interdisciplinary integration. Despite the existence of current databases of scientific literature such as Scopus, Web of Science, and Crossref, which offer ways to search for research papers based on specific labels, extracting useful information of material science from the vast ocean of literature remains demanding. To obtain clearer sense of materials properties some structured database project such as arXiv:2404.03080v1 [cs.CL] 3 Apr 2024 Materials Project[2], OQMD[3], and NOMAD[4] were developed. However, these databases contain many computational results obtained through techniques like Density Functional Theory (DFT) or Molecular Dynamics (MD) simulations[5]. While these computational databases can provide valuable references for predicting and understanding certain materials systems, they often face discrepancies with experimental observations. Therefore, there is an urgent need within the field of materials science for a database grounded in experimental research and practical information. Furthermore, the complexity of materials information extends beyond composition and structure to encompass their respective application fields. For instance, organic materials are commonly utilized in biological applications, semiconductors serve as integral components in electronics, and metals find applications in mechanical engineering. The process of designing novel materials typically commences with a clear understanding of their intended applications, aiming to maximize research efficiency. Given this context, compared to universal databases like the Materials Project, which might not be as useful due to their broad focus, specific databases focused on applications or traits hold the potential to provide more valuable information for researchers in relevant industries. This underscores the importance of developing dedicated databases that cater to the nuanced needs of materials science research, facilitating a more targeted approach to material discovery and application. Knowledge graph (KG) is a structured representation of information that models the controlled vocabulary and ontological relations of a topical domain as nodes and edges, enabling complex queries and insights that traditional databases cannot easily provide. The adoption of knowledge graphs offers several advantages, including enhanced data interoperability, the ability to infer new knowledge through relational data analysis, and improved data quality and consistency through structured representation[6, 7]. These features make knowledge graphs particularly valuable for integrating diverse information sources and providing a unified view of a domain’s knowledge, thereby facilitating more informed decision-making and discovery[8]. However, the construction of knowledge graphs in specific fields always requires the participation of a large number of expert[9]. This labor-intensive process not only limits the scalability of KGs but also impacts their performance and timeliness10. With the rapid development of natural language processing (NLP), methods for extracting information from unstructured text and constructing knowledge graphs have become more efficient and accurate[11]. For instance, in 2016, the Metallic Materials Knowledge Graph (MMKG) was developed to store materials information from various web data resources[12]. Knowledge graphs tailored to lithium-ion battery cathodes have been constructed, aimed at identifying potential new materials candidates[13]. A user-friendly databases focusing on specific material types, such as Metal-Organic Framework Knowledge Graphs (MOF-KG), have been developed[14]. Recently, a material knowledge graph, MatKG and MatKG2, containing information on material properties, structure, and applications, has been developed[1, 15]. However, these material knowledge graphs face even greater challenges. Firstly, although advancements in NLP technology have reduced the dependency on experts to a certain extent, training data still requires extensive annotation to enhance model accuracy[16]. Secondly, the construction of these knowledge graphs often involves predicting relationships between nodes to form triples, which means the entities represented in the KG are not always based on real instances[17]. This can diminish the authenticity and credibility of the KG. Additionally, this approach makes updating the knowledge graph difficult, as each new node introduced necessitates predicting its relationship with every other node, complicating the maintenance of a dynamic and accurate knowledge graph, especially in advanced fields like material science. Acknowledging these challenges, the emergence of LLMs like GPT and LLaMA represents a breakthrough, offering new solutions to enhance the extraction and credibility of structured information[18], 19. The fine-tuning technique of LLMs can significantly enhance their performance in specific domain text tasks through training with fewer samples[20 , 21]. This means improving the results of NER and RE without requiring a large amount of labor becomes possible and was adopt in our research. In this paper, we have achieved significant advancements in the development of Functional Materials Knowledge Graph (FMKG), a pioneering graph database tailored for the field of functional materials. Our contributions are highlighted in three key areas: 1) We propose a method to achieve named entity recognition (NER), relation extraction (RE) and entity resolution (ER) with high accuracy. Through this method, we can easily convert unstructured text into triples and retain the source information of each triplet. This method also makes updating KG very convenient. 2) We constructed the first knowledge graph dedicated for functional materials, where researchers can easily get the information about functional material through query the FMKG. 3) We use a well-defined label system so our KG can be easily scale-up and potentially combined with other structured databases or KGs. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Yanpeng Ye, School of Computer Science and Engineering, University of New South Wales, Kensington, NSW, Australia, GreenDynamics Pty. Ltd, Kensington, NSW, Australia, and these authors contributed equally to this work; (2) Jie Ren, GreenDynamics Pty. Ltd, Kensington, NSW, Australia, Department of Materials Science and Engineering, City University of Hong Kong, Hong Kong, China, and these authors contributed equally to this work; (3) Shaozhou Wang, GreenDynamics Pty. Ltd, Kensington, NSW, Australia (shaozhou@greendynamics.com.au); (4) Yuwei Wan, GreenDynamics Pty. Ltd, Kensington, NSW, Australia and Department of Linguistics and Translation, City University of Hong Kong, Hong Kong, China; (5) Imran Razzak, School of Computer Science and Engineering, University of New South Wales, Kensington, NSW, Australia; (6) Tong Xie, GreenDynamics Pty. Ltd, Kensington, NSW, Australia and School of Photovoltaic and Renewable Energy Engineering, University of New South Wales, Kensington, NSW, Australia (tong@greendynamics.com.au); (7) Wenjie Zhang, School of Computer Science and Engineering, University of New South Wales, Kensington, NSW, Australia (wenjie.zhang@unsw.edu.au). Authors: Authors: (1) Yanpeng Ye, School of Computer Science and Engineering, University of New South Wales, Kensington, NSW, Australia, GreenDynamics Pty. Ltd, Kensington, NSW, Australia, and these authors contributed equally to this work; (2) Jie Ren, GreenDynamics Pty. Ltd, Kensington, NSW, Australia, Department of Materials Science and Engineering, City University of Hong Kong, Hong Kong, China, and these authors contributed equally to this work; (3) Shaozhou Wang, GreenDynamics Pty. Ltd, Kensington, NSW, Australia (shaozhou@greendynamics.com.au); (4) Yuwei Wan, GreenDynamics Pty. Ltd, Kensington, NSW, Australia and Department of Linguistics and Translation, City University of Hong Kong, Hong Kong, China; (5) Imran Razzak, School of Computer Science and Engineering, University of New South Wales, Kensington, NSW, Australia; (6) Tong Xie, GreenDynamics Pty. Ltd, Kensington, NSW, Australia and School of Photovoltaic and Renewable Energy Engineering, University of New South Wales, Kensington, NSW, Australia (tong@greendynamics.com.au); (7) Wenjie Zhang, School of Computer Science and Engineering, University of New South Wales, Kensington, NSW, Australia (wenjie.zhang@unsw.edu.au). Editor’s note: This article is part of a broader study. You’re reading Part 1 of 9. Read the rest below. Editor’s note: This article is part of a broader study. You’re reading Part 1 of 9. Read the rest below. Editor’s note: This article is part of a broader study. You’re reading Part 1 of 9. Read the rest below. Table of Links Abstract and Introduction Methods Data preparation and schema design LLMs training, evaluation and inference Entity resolution Knowledge graph construction Result Discussion Conclusion and References Abstract and Introduction Abstract and Introduction Methods Methods Data preparation and schema design Data preparation and schema design LLMs training, evaluation and inference LLMs training, evaluation and inference Entity resolution Entity resolution Knowledge graph construction Knowledge graph construction Result Result Discussion Discussion Conclusion and References Conclusion and References ABSTRACT The convergence of materials science and artificial intelligence has unlocked new opportunities for gathering, analyzing, and generating novel materials sourced from extensive scientific literature. Despite the potential benefits, persistent challenges such as manual annotation, precise extraction, and traceability issues remain. Large language models have emerged as promising solutions to address these obstacles. This paper introduces Functional Materials Knowledge Graph (FMKG), a multidisciplinary materials science knowledge graph. Through the utilization of advanced natural language processing techniques, extracting millions of entities to form triples from a corpus comprising all high-quality research papers published in the last decade. It organizes unstructured information into nine distinct labels, covering Name, Formula, Acronym, Structure/Phase, Properties, Descriptor, Synthesis, Characterization Method, Application, and Domain, seamlessly integrating papers’ Digital Object Identifiers. As the latest structured database for functional materials, FMKG acts as a powerful catalyst for expediting the development of functional materials and a fundation for building a more comprehensive material knowledge graph using full paper text. Furthermore, our research lays the groundwork for practical text-mining-based knowledge management systems, not only in intricate materials systems but also applicable to other specialized domains. Introduction In the contemporary information era, despite notable advancements, the creation and advancement of novel materials still heavily rely on traditional, time-consuming trial-and-error methods intertwined with chemical and physical intuitions. These conventional research approaches significantly impede the life-cycle of high-performance material research. Given the specialization, inherent complexity, and vast knowledge base of material science, researchers focusing on a single direction often struggle to efficiently access and understand material knowledge from multidisciplinary studies. For instance, researchers in solar cell development might not fully comprehend studies related to solid-state batteries or organic light-emitting diodes. Yet, the electronic properties of materials across these different domains are highly related, and researchers in different domains can potentially learn from each other. To accelerate the progress of materials research, there is a pressing need to efficiently integrate knowledge from various disciplines[1]. However, this vital knowledge is scattered across a vast array of over 10 million scientific papers, covering diverse topics and disciplines such as materials preparation and functionalization methods, advanced materials characterization techniques, and the exploration of physical, chemical, and biological properties, along with their applications in fields like electronic devices, clean energy storage and transfer, and mechanical engineering. This fragmentation of knowledge represents a significant barrier to interdisciplinary collaboration and innovation. A critical gap in current research infrastructure is the lack of an effective materials science database that can consolidate this scattered knowledge, facilitating easier access and interdisciplinary integration. Despite the existence of current databases of scientific literature such as Scopus, Web of Science, and Crossref, which offer ways to search for research papers based on specific labels, extracting useful information of material science from the vast ocean of literature remains demanding. To obtain clearer sense of materials properties some structured database project such as arXiv:2404.03080v1 [cs.CL] 3 Apr 2024 Materials Project[2], OQMD[3], and NOMAD[4] were developed. However, these databases contain many computational results obtained through techniques like Density Functional Theory (DFT) or Molecular Dynamics (MD) simulations[5]. While these computational databases can provide valuable references for predicting and understanding certain materials systems, they often face discrepancies with experimental observations. Therefore, there is an urgent need within the field of materials science for a database grounded in experimental research and practical information. Furthermore, the complexity of materials information extends beyond composition and structure to encompass their respective application fields. For instance, organic materials are commonly utilized in biological applications, semiconductors serve as integral components in electronics, and metals find applications in mechanical engineering. The process of designing novel materials typically commences with a clear understanding of their intended applications, aiming to maximize research efficiency. Given this context, compared to universal databases like the Materials Project, which might not be as useful due to their broad focus, specific databases focused on applications or traits hold the potential to provide more valuable information for researchers in relevant industries. This underscores the importance of developing dedicated databases that cater to the nuanced needs of materials science research, facilitating a more targeted approach to material discovery and application. Knowledge graph (KG) is a structured representation of information that models the controlled vocabulary and ontological relations of a topical domain as nodes and edges, enabling complex queries and insights that traditional databases cannot easily provide. The adoption of knowledge graphs offers several advantages, including enhanced data interoperability, the ability to infer new knowledge through relational data analysis, and improved data quality and consistency through structured representation[6, 7]. These features make knowledge graphs particularly valuable for integrating diverse information sources and providing a unified view of a domain’s knowledge, thereby facilitating more informed decision-making and discovery[8]. However, the construction of knowledge graphs in specific fields always requires the participation of a large number of expert[9]. This labor-intensive process not only limits the scalability of KGs but also impacts their performance and timeliness10. With the rapid development of natural language processing (NLP), methods for extracting information from unstructured text and constructing knowledge graphs have become more efficient and accurate[11]. For instance, in 2016, the Metallic Materials Knowledge Graph (MMKG) was developed to store materials information from various web data resources[12]. Knowledge graphs tailored to lithium-ion battery cathodes have been constructed, aimed at identifying potential new materials candidates[13]. A user-friendly databases focusing on specific material types, such as Metal-Organic Framework Knowledge Graphs (MOF-KG), have been developed[14]. Recently, a material knowledge graph, MatKG and MatKG2, containing information on material properties, structure, and applications, has been developed[1, 15]. However, these material knowledge graphs face even greater challenges. Firstly, although advancements in NLP technology have reduced the dependency on experts to a certain extent, training data still requires extensive annotation to enhance model accuracy[16]. Secondly, the construction of these knowledge graphs often involves predicting relationships between nodes to form triples, which means the entities represented in the KG are not always based on real instances[17]. This can diminish the authenticity and credibility of the KG. Additionally, this approach makes updating the knowledge graph difficult, as each new node introduced necessitates predicting its relationship with every other node, complicating the maintenance of a dynamic and accurate knowledge graph, especially in advanced fields like material science. Acknowledging these challenges, the emergence of LLMs like GPT and LLaMA represents a breakthrough, offering new solutions to enhance the extraction and credibility of structured information[18], 19. The fine-tuning technique of LLMs can significantly enhance their performance in specific domain text tasks through training with fewer samples[20 , 21]. This means improving the results of NER and RE without requiring a large amount of labor becomes possible and was adopt in our research. In this paper, we have achieved significant advancements in the development of Functional Materials Knowledge Graph (FMKG), a pioneering graph database tailored for the field of functional materials. Our contributions are highlighted in three key areas: 1) We propose a method to achieve named entity recognition (NER), relation extraction (RE) and entity resolution (ER) with high accuracy. Through this method, we can easily convert unstructured text into triples and retain the source information of each triplet. This method also makes updating KG very convenient. 2) We constructed the first knowledge graph dedicated for functional materials, where researchers can easily get the information about functional material through query the FMKG. 3) We use a well-defined label system so our KG can be easily scale-up and potentially combined with other structured databases or KGs. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv