Eliot: Interactively $\underline{E}$xploring Fast-Changing Scientific $\underline{Li}$terature Trends with $\underline{O}$nline Da$\underline{t}$a and Learning

Researchers present Eliot, an interactive system for exploring evolving scientific literature trends across rapidly changing fields like Large Language Models and Automated Planning. The tool retrieves arXiv papers at query time, clusters them into thematic groups, and visualizes publication patterns over time, with evaluations showing 85% accuracy in meaningful cluster labeling across eight research domains.

Eliot addresses a critical challenge in modern scientific research: the exponential growth of publications has made it nearly impossible for researchers to track trends in fast-moving fields manually. The system combines information retrieval with machine learning clustering to provide traceable, auditable explorations of literature evolution, moving beyond opaque search engine results and LLM summaries that obscure how conclusions were derived.

The development stems from observed limitations in existing tools. Search engines and AI assistants retrieve relevant papers but rarely expose their selection methodology or temporal patterns, making it difficult for researchers to verify claims or understand field trajectories. Eliot solves this by performing real-time clustering of arXiv papers using MiniLM embeddings with 10-dimensional UMAP reduction and Agglomerative Clustering—configurations validated across multiple domains.

The system's practical value manifests in its interpretability metrics. User studies demonstrated that researchers found cluster labels meaningful 85% of the time and valued Eliot most for creating auditable overviews of rapidly evolving technical areas. This capability proves especially valuable for researchers entering new subfields, competitive analysis, and institutional literature reviews where transparency matters. The query-time clustering approach preserves the freshness advantage of current search while adding structural understanding that pure retrieval cannot provide.

Looking forward, Eliot represents a broader shift toward explainable AI tools that complement rather than replace human judgment. As scientific output accelerates, systems combining clustering, temporal visualization, and auditable methodology will become essential infrastructure for researchers. The tool's deployment signals growing recognition that discoverability requires both relevance and transparency.

• →Eliot enables interactive, real-time exploration of scientific literature trends with transparent corpus selection and clustering methodologies.
• →Evaluations across eight arXiv domains validate MiniLM embeddings combined with UMAP dimensionality reduction and Agglomerative Clustering as optimal configurations.
• →User studies confirmed 85% accuracy in cluster meaningfulness, with researchers valuing the system for auditable overviews of rapidly changing technical fields.
• →The system complements search and LLM tools by providing traceable evidence inspection rather than opaque summarization or retrieval.
• →Query-time clustering maintains currency while adding structural understanding that helps researchers verify field trends and understand publication patterns.