An AI system that incrementally generates scientific paper drafts by predicting links between concepts and generating text sections.
PaperRobot is an AI-powered system that helps researchers generate draft content for scientific papers. It uses knowledge graphs built from PubMed literature to predict relationships between concepts and incrementally produce coherent text sections like titles, abstracts, and conclusions. The system addresses the challenge of writer's block in academic writing by providing AI-assisted idea exploration and organization.
Computational linguistics researchers, AI scientists working on text generation, and academic researchers looking for AI tools to assist with scientific writing and idea development.
PaperRobot offers a unique incremental approach to scientific writing that combines knowledge graph reasoning with text generation, unlike generic language models. It's specifically trained on PubMed data for scientific domains and provides structured generation of paper sections rather than free-form text.
Code for PaperRobot: Incremental Draft Generation of Scientific Ideas
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Builds scientific paper sections step-by-step, mimicking human writing processes as described in the key features, which helps overcome writer's block.
Uses datasets from PubMed, making it highly tailored for biomedical research text generation, unlike generic language models.
Predicts relationships between scientific concepts using link prediction, enhancing text coherence and idea exploration based on the architecture.
Accepted at ACL 2019, indicating peer-reviewed quality and a research-focused approach, as cited in the README.
Requires downloading large datasets, specific Python 3.6 and Ubuntu versions, and manual training steps, as detailed in the Quickstart section.
The README cautions that training and evaluation are 'pretty slow' due to the dataset size, limiting real-time use.
Trained exclusively on PubMed data, so it may not generalize well to other scientific fields or non-scientific text generation.
Relies on Pytorch 1.1 and other packages from 2019, which could have compatibility issues with modern systems and lack ongoing support.