EPSCoR Research Fellows: NSF: Deep Generative Models for Analysis and Visualization of Scientific Texts

The number of scientific publications is continually increasing, which poses challenges for researchers in digesting and discovering relevant insights from work within their discipline and across multiple disciplines. To help researchers overcome information overload and keep up with the latest research, the overall goal of this project is to support the scientific literature review process and facilitate creative scientific problem-solving by developing models and algorithms for the analysis and visualization of scientific texts.

The project will provide a fellowship to an Assistant professor and training for a graduate student at the New Mexico State University. This work will be conducted in collaboration with researchers at the University of California, San Diego (UCSD). Through the fellowship, the principal investigator aims to create algorithms that model the intent, topics, and embeddings within scientific documents.

Additionally, the PI, in partnership with UCSD, will develop an interactive visual tool that enables users to explore related articles based on the document's intent structure. This project will benefit the community by providing an efficient tool for researchers to browse, search, and manage research articles, enabling them to extract useful insights and relationships between articles, which will hopefully accelerate scientific progress.

The project will contribute to text mining research by introducing new approaches for mining and visualizing scientific texts. More specifically, this project aims to (i) develop neural topic models for supervised document structure learning and intent structure comparison of scientific articles; (ii) create semantic visualization methods that jointly model textual content, rhetorical structure, and topic structure of documents for exploratory tasks such as browsing or finding relevant articles; and (iii) build an interactive visual tool for mining scientific texts based on the proposed models.

This project has strong potential to advance the theory and practice of text mining, document modeling, and visualization The project's proposed methods and interactive visual tool can be integrated into computational systems that could help accelerate scientific progress by assisting researchers in managing information overload.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.