A new AI tool builds very good knowledge graphics, and can re -connect the scientific discovery
Authors:
(1) Yanpeng Ye, College of Computer Science and Engineering, New South Wales University, Kinsington, New South Wales, Australia, Grendynamics Pty. LTD, Kensington, NSW, Australia, and these authors contributed equally to this work;
(2) Jie Ren, Grendynamics Pty. LTD, Kensington, NSW, Australia, Department of Materials Science and Engineering, Hong Kong University, Hong Kong, China, and these authors contributed equally to this work;
(3) Shaozhou Wang, Grendynamics Pty. LTD, Kensington, NSW, Australia ([email protected]);
(4) Yuwei Wan, Grendynamics Pty. LTD, Kensington, NSW, Australia and Linguistics and Translation Department, Hong Kong University, Hong Kong, China;
(5) Imran Razak, College of Computer Science and Engineering, New South Wales University, Kensington, New South Wales, Australia;
(6) Tong XIE, Grendynamics Pty. LTD, Kensington, NSW, Australia and School of Environment and Renewal Energy Engineering, New South Wales University, Kinsington, New South Wales, Australia ([email protected]);
(7) Winji Chang, College of Computer Science and Engineering, New South Wales University, Kinsington, New South Wales, Australia ([email protected]).
In this study, we offer a new NLP pipeline for the construction of KG, which aims to extract three times from the non -structured scientific texts. The main advantage of the method is that it can adjust LLMS by commenting on a small amount of data, and using the seized LLM to extract the information organized from a large amount of non -structured text. The entire process does not depend on any prediction, which can increase the authenticity of the information organized and follow it to the maximum. By using this method, we build a graph of functional knowledge (FMKG) that contains materials and their relevant knowledge from the summary of 150,000 sheets that the pendant reviewed. After the analysis, we showed the effectiveness and credibility of FMKG.
In addition, our way and KG have great potential in different dimensions. First, enhancing the depth of extracting structured information to include entire research papers with a more richer and more detailed knowledge. Not only does this involve expanding the scope of the data that has been analyzed, but also improves the process to capture the nuances in complex scientific texts. Second, the refining of entity stickers within our system allows a more accurate classification of data, including combining detailed features such as synthesis conditions or characteristics, which greatly improves details and the benefit of the graph of knowledge. Third, the ingenuity of our NLP pipeline indicates its application through various scientific fields, which provides a template to create the knowledge graphs of the field beyond material science. Finally, the FMKG integration with current knowledge charts such as MATKG opens ways to create a more connected and comprehensive data collection, and facilitate the development of advanced research and the development of applications in material science and beyond.
Venugupal, V. & Olivetti, E. MATKG: a graphic drawing that was created independently in material science. Sci. Data 11, 217 (2024).
Jain, a. And others. Articles Project: Genome approach material to accelerate material innovation, APL MATER. APL materials (2013).
SAAL, Je, Kirklin, S., Aykol, M., Meredig, B. & Wolverton, C. Design and Discovery with a high density functional theory: Open Quantum database (OQMD). JOM 65, 1501-1509 (2013).
Draxl, C. & Scheffler, M. The Nomad Laboratory: From data exchange to artificial intelligence. J. Physics: Matter. 2, 036001 (2019).
Mrdjenovich, D. et al. ProPNet: A graphic drawing of material science. Article 2, 464-480 (2020).
Ji, S., Pan, S., Cambria, E., MartTinen, P. & Yu, PS A Quorly on Prackts Knowledge: Acting, Size, and Applications. IEEE transactions on nervous networks learn. split. 33, 494-514 (2022).
Chang, c. AI Open 2, 14-35 (2021).
Mitchell, T. And others. The endless learning. communication. ACM 61, 103-115 (2018).
Zhong, L., Wu, J., LI, Q. ACM Comput. Survival. 56 (2023).
Ban, S. And others. Standardizing the large language models and graphic fees for knowledge: a road map. IEEE transactions on Knowl. Engineer Data. 1-20 (2024).
Weston, l. And others. The entity’s recognition and normalization of information is widely applied from material science literature has been applied. J. Chem. Inf. model. 59, 3692-3702 (2019).
Zhang, X., LIU, X., LI, X. & PAN, D. MMKG: A way to create a graph for metal materials on the basis of DBIDIA and Wikipedia. account. physics. communication. 211, 98-112 (2017).
Ni, g. And others. Automation of material exploring with a graphic fee for the semantic knowledge of the Li-IIN battery cathode. ADV. FUNCT. the mom. 32, 2201437 (2022).
An, y. et al. Question of the graph for knowledge answering material science (KGQA4MAT): Development of the natural language interface for metal metal frameworks (MOF-KG). Arxiv Preprint Arxiv: 2309.11361 (2023).
Venugupal, V. & Olivetti, E. MATKG-2: Unveiled the science of scientific existence of microscopic materials through independent committees of LLMS. Amnesty International for the series. the mom. Des. – Neups 2023 work. (2023).
SU, P., LI, G., Wu, C. & Vijay-Shanker, K. Using the long supervision of manually expressed data to extract the relationship. Plos One 14 (2019).
Sousa, RT, Silva, S. & Pesquita, C. Interpreted representations to predict the relationship in knowledge graphics. Arxiv Preprint Arxiv: 2306.12687 (2023).
Brown, T. And others. Language models are few learners. ADV. Neurological information processing systems 33, 1877-1901 (2020).
Touvron, H. Et al. Lama: Open and effective foundation models. Arxiv Preprint Arxiv: 2302.13971 (2023).
Xie, T. et al. Create an organized and paid solar cell data collection with large language models. Patterns (2024).
Dagdelen, J. et al. Extracting information from the scientific text with large language models. Nat. communication. 15, 1418 (2024).
Swain, MC & Cole, JM ChemdataExtractor: Automated Tools for Chemical Information Tools from Scientific Literature. J. Chem. Inf. model. 56, 1894-1904 (2016).
Tshitoyan, V. et al. Words that are not subject to supervision implications, capture the underlying knowledge of material science literature. Nature 571, 95+ (2019).
Xie, T. et al. Darwin Series: Large Language Models for Natural Sciences. Arxiv Preprint Arxiv: 2308.13565 (2023).
