AI BriefingLG AI Research 272
Key point
LG AI Research collaborated with Elsevier to achieve the top performance in the ChEMU 2022 challenge, which extracts reaction information from chemical patents and journals.
Details
LG AI Research is researching Information Extraction technology in the chemistry field in collaboration with the world-renowned scientific publisher Elsevier. Information extraction is a technology that automatically extracts structured information from raw text, and it plays a core role in the Reaxys service, which systematizes chemical reaction and material information.
LG AI Research participated in the ChEMU 2022 challenge held at CLEF in 2022, and recorded the highest performance among participating teams in two tasks (1a, 1b) that extract chemical reaction information from chemical patents.
The main tasks it participated in are as follows:
- Task 1a (Named Entity Recognition): Extracting entities describing chemical reactions, such as chemical substances and numerical information
- Task 1b (Event Extraction): Extracting trigger words describing chemical reactions and the relations between entities
To improve performance, LG AI Research utilized a domain-specific language model based on the Bio-LinkBert model. The model was post-trained using vast amounts of data, including chemical patents (23GB), journals (22GB), and Pubmed (20GB), and it was confirmed that the highest performance was achieved when trained on a combination of chemical journal and Pubmed data in particular.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.