INTELLIGENT RESEARCH NAVIGATOR: AN AI DRIVEN FRAMEWORK FOR AUTONOMOUS INNOVATION AND DISCOVERY

Abstract

Research has developed at a rapid rate, which has created an overwhelming amount of literature for scientists to review before proceeding with any new research - this is especially true when identifying new areas of study that have not been conducted. An intelligent research navigator (IRN) is a fully automated process for identifying unexplored research areas and developing research ideas. The IRN combines several methods including using the strengths of large language model reasoning and creating a combination of document clusters using the density-based clustering technique, and analysing longitudinal publication trends to generate a citation network. The IRN contains five separate agents that work together to provide researchers with actionable insights. An example of these agents would be the query-understanding agent, which retrieves papers from various databases, the semantic clustering agent (groups together papers that feel similar) uses the information to produce a set of gaps and prioritize them for researchers, and the ideas synthesising agent, which uses the gaps as input to generate new ideas. The first five scientific domains were within 2 minutes of providing researchers with actionable insights, and over 50% of the gaps that were generated by the system would not have been identified through traditional systematic review methods. Similarly, 80% of the research ideas generated by the system were considered valid by experts in those fields. Validation of the scoring process was confirmed by the high level of rank correlation between the automated quality scores and the quality assessments produced by expert rater (Spearman's ρ = 0.71).