Large-Scale Multidimensional Knowledge Profiling of Scientific Literature

TL;DR

The paper tackles the difficulty of synthesizing rapidly growing AI literature by building a large-scale, semantically grounded profiling pipeline that processes over 100,000 papers from 22 major conferences (2020–2025) into a structured ResearchDB. It combines semantic parsing with topic clustering and hierarchical retrieval to enable multidimensional analyses of topics, datasets, models, and institutional patterns. Key findings include a shift toward safety, multimodal reasoning, and agent-oriented research, alongside compute and dataset trends and a core-multipoint collaboration network among leading institutions. The framework supports evidence-based trend analysis and decision-making for researchers, funders, and industry, while providing an extensible resource for future meta-analyses.

Abstract

The rapid expansion of research across machine learning, vision, and language has produced a volume of publications that is increasingly difficult to synthesize. Traditional bibliometric tools rely mainly on metadata and offer limited visibility into the semantic content of papers, making it hard to track how research themes evolve over time or how different areas influence one another. To obtain a clearer picture of recent developments, we compile a unified corpus of more than 100,000 papers from 22 major conferences between 2020 and 2025 and construct a multidimensional profiling pipeline to organize and analyze their textual content. By combining topic clustering, LLM-assisted parsing, and structured retrieval, we derive a comprehensive representation of research activity that supports the study of topic lifecycles, methodological transitions, dataset and model usage patterns, and institutional research directions. Our analysis highlights several notable shifts, including the growth of safety, multimodal reasoning, and agent-oriented studies, as well as the gradual stabilization of areas such as neural machine translation and graph-based methods. These findings provide an evidence-based view of how AI research is evolving and offer a resource for understanding broader trends and identifying emerging directions. Code and dataset: https://github.com/xzc-zju/Profiling_Scientific_Literature

Figures (12)

• Figure 1: Number of papers published and topic statistics across 22 conferences from 2020 to 2025.
• Figure 2: LLM-Driven Multi-Dimensional Knowledge Profiling Framework with Large-Scale Literature Database and Hierarchical Retrieval.
• Figure 3: Four-quadrant topic lifecycle evolution (quadrants I-IV: Emerging, Booming, Mature, Declining). The inset (bottom left) shows the most prolific topic by paper count from 2020-2025.
• Figure 4: Compute Resource Analysis. Left: Temporal evolution of compute demand; Right: Boxplot of the top 10 topics with highest compute requirements.
• Figure 5: Dataset and Benchmark Trends (2020-2025). The bar chart shows annual paper counts for each dataset type, and the line chart shows their proportion of total publications.