CollabNext Team 20: Knowledge Graph Q&A System for Research

This system lets users ask questions in plain language about researchers, topics, and institutions in the OpenAlex research database. It focuses on illuminating diversity by highlighting underrepresented contributors in science, starting with researchers at HBCUs including Georgia Tech, Morehouse College, Clark Atlanta University, and University of Buffalo. The system uses Retrieval Augmented Generation to answer questions with exact insights from the knowledge graph, reducing hallucinations from generative language models. It calculates centrality and community information to help discover underrepresented researchers whose work might otherwise be hard to find.

The team built a functional system with three main components: an ETL pipeline that extracts OpenAlex data and loads it into Neo4j, augmentation that adds centrality and community labels to the data, and a web UI retrieval app that answers user questions with graph insights. The generic ETL design uses only about 600 lines of code and defines labels and properties of interest simply, making it easily adaptable to other datasets like clinical trials. The approach combines the OpenAlex graph-based research database with modern language model chatbots.

The system makes research insights accessible to users without technical knowledge of Cypher query languages. Previously, fully utilizing OpenAlex required programming skills. Primary beneficiaries include Georgia Tech, Morehouse College, Clark Atlanta University, University of Buffalo, independent researchers, and companies. The prototype successfully answered target personas' questions that would normally require technical skills.