Author:
Xiao Zhengyang,Pakrasi Himadri B.,Chen Yixin,Tang Yinjie J.
Abstract
AbstractLarge language models (LLMs) can complete general scientific question-and-answer, yet they are constrained by their pretraining cut-off dates and lack the ability to provide specific, cited scientific knowledge. Here, we introduceNetwork forKnowledgeOrganization (NEKO), a workflow that uses LLM Qwen to extract knowledge through scientific literature text mining. When user inputs a keyword of interest, NEKO can generate knowledge graphs and comprehensive summaries from PubMed search. NEKO has immediate applications in daily academic tasks such as education of young scientists, literature review, paper writing, experiment planning/troubleshooting, and new hypothesis generation. We exemplified this workflow’s applicability through several case studies on yeast fermentation and cyanobacterial biorefinery. NEKO’s output is more informative, specific, and actionable than GPT-4’s zero-shot Q&A. NEKO offers flexible, lightweight local deployment options. NEKO democratizes artificial intelligence (AI) tools, making scientific foundation model more accessible to researchers without excessive computational power.
Publisher
Cold Spring Harbor Laboratory
Reference26 articles.
1. Artificial intelligence: a solution to involution of design–build–test–learn cycle
2. OpenAI. GPT-4 Technical Report. arXiv preprint 2303.08774 (2023).
3. Qwen technical report;arXiv preprint,2023
4. When Do LLMs Need Retrieval Augmentation? Mitigating LLMs’ Overconfidence Helps Retrieval Augmentation;arXiv preprint,2024
5. Retrieval-augmented generation for knowledge-intensive nlp tasks;Advances in Neural Information Processing Systems,2020