Mapping Literature Landscapes with Data-Driven Discovery: A Case Study on MOEA/D

Mingyu Huang, Ke Li, Li, Ke

Abstract

We are living in an era of "big literature", where scientific literature is expanding exponentially. While this growth presents new opportunities, it complicates mapping global scientific research landscapes, as manual review methods become infeasible. Recent advancements in machine learning, complex networks, and natural language processing have enabled numerous data-driven discovery methods. Building upon these tools, we introduce an end-to-end workflow for analyzing large-scale literature landscapes, LitLA. This workflow first integrates diverse publication metadata into a bibliographic knowledge graph (KG) representing the research landscape. It then offers tools for exploratory analysis of various landscape aspects. We demonstrate the effectiveness of LitLA via a case study on follow-up works of multi-objective evolutionary algorithm based on decomposition (MOEA/D). In doing so, we constructed the MOEA/D research landscape as a KG comprising over 5,400 papers, 10,000 authors, 1,600 institutions, and 78,000 keywords. With this landscape, we start with descriptive statistics and investigate prominent topics pertaining to MOEA/D and interrogate their spatial-temporal and bilateral relationships. We then map the collaboration and citation networks to reveal the community's growth over time. We further experiment whether learning on latent patterns of this landscape can hint on future research directions.

Introduction

Scientific output is expanding at an exponential rate, leading to a phenomenon known as "Big Literature." This explosion of data makes it increasingly difficult for researchers to maintain a comprehensive overview of their fields. Traditional Systematic Literature Reviews (SLRs), while rigorous, are manually intensive and prone to selection bias. In fast-moving domains, a standard survey might only capture a fraction of relevant work, leaving thousands of papers unexamined. To address this, we propose LitLA (Literature Landscape Analysis), an automated, data-driven workflow designed to synthesize the landscape of a research area objectively.

Methodology: The LitLA Workflow

LitLA transforms raw publication metadata into a structured, actionable Knowledge Graph (KG). The workflow consists of four primary stages:

  1. Knowledge Graph Construction: Building a heterogeneous network where nodes represent papers, authors, institutions, venues, and keywords, capturing the complex web of relationships within a field.
  2. Topic Modeling: Utilizing deep learning embeddings (Voyage AI) and clustering (HDBSCAN) via BERTopic to identify distinct research themes. GPT-4o is employed to generate human-readable summaries for these clusters.
  3. Network Science: Analyzing citation networks to identify the field's "backbone" and collaboration networks to uncover influential hubs and "small-world" effects.
  4. Future Trend Forecasting: Treating research evolution as a link-prediction task. By analyzing historical keyword co-occurrences, a gradient-boosting model predicts likely future conceptual merges.
The LitLA workflow: an end-to-end process from database search and metadata collection to network analysis and trend forecasting.

Case Study: The MOEA/D Research Landscape

We applied LitLA to the field of Multi-Objective Evolutionary Algorithms based on Decomposition (MOEA/D). Since its inception in 2007, this domain has grown from a single seminal paper to a massive landscape. Our dataset, sourced from the Web of Science (WoS) and enriched with Semantic Scholar S2ORC citation contexts, includes 5,404 papers, 10,532 researchers, and nearly 80,000 keywords.

Key Results and Insights

Our analysis reveals a field expanding at a staggering annual growth rate of 40.5%. While the early years were dominated by engineering applications, Computer Science—specifically Machine Learning—has become the fastest-growing category since 2012.

General statistics of the MOEA/D landscape, illustrating the exponential growth in publications (A) and authors (B), alongside a breakdown of dominant research categories (C).

Key findings include:

  • Structural Topology: Topic visualization shows a central "continent" of methodological research surrounded by "islands" of specialized applications like scheduling and energy management.
  • Maturation Trends: The "CD index" suggests declining disruptiveness over time, as the field shifts from paradigm-shifting work to the consolidation and refinement of existing knowledge.
  • Predictive Success: The link-prediction model successfully forecasted the 2024 surge in research at the intersection of Large Language Models (LLMs) and evolutionary optimization.

Conclusion

LitLA demonstrates that automated, data-driven discovery can effectively map vast research landscapes that are otherwise impenetrable to manual review. While the current implementation relies on citation-based selection and specific topic modeling hyperparameters, future iterations will integrate deeper LLM-based contextualization to explain the underlying drivers of research shifts in real-time.

Summary updated

Abstract

We are living in an era of "big literature", where scientific literature is expanding exponentially. While this growth presents new opportunities, it complicates mapping global scientific research landscapes, as manual review methods become infeasible. Recent advancements in machine learning, complex networks, and natural language processing have enabled numerous data-driven discovery methods. Building upon these tools, we introduce an end-to-end workflow for analyzing large-scale literature landscapes, LitLA. This workflow first integrates diverse publication metadata into a bibliographic knowledge graph (KG) representing the research landscape. It then offers tools for exploratory analysis of various landscape aspects. We demonstrate the effectiveness of LitLA via a case study on follow-up works of multi-objective evolutionary algorithm based on decomposition (MOEA/D). In doing so, we constructed the MOEA/D research landscape as a KG comprising over 5,400 papers, 10,000 authors, 1,600 institutions, and 78,000 keywords. With this landscape, we start with descriptive statistics and investigate prominent topics pertaining to MOEA/D and interrogate their spatial-temporal and bilateral relationships. We then map the collaboration and citation networks to reveal the community's growth over time. We further experiment whether learning on latent patterns of this landscape can hint on future research directions.
BibTeX
@misc{huang2024mapping,
  author = {Mingyu Huang and Ke Li and Li, Ke},
  title = {{Mapping Literature Landscapes with Data-Driven Discovery: A Case Study on MOEA/D}},
  year = {2024},
  eprint = {2404.14228},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2404.14228},
}
Figure