A bibliometric approach to Systematic Mapping Studies: The case of the evolution and perspectives of community detection in complex networks

Critical analysis of the state of the art is a necessary task when identifying new research lines worthwhile to pursue. To such an end, all the available work related to the field of interest must be taken into account. The key point is how to organize, analyze, and make sense of the huge amount of scientific literature available today on any topic. To tackle this problem, we present here a bibliometric approach to Systematic Mapping Studies (SMS). Thus, a modify SMS protocol is used relying on the scientific references metadata to extract, process and interpret the wealth of information contained in nowadays research literature. As a test case, the procedure is applied to determine the current state and perspectives of community detection in complex networks. Our results show that community detection is a still active, far from exhausted, in development, field. In addition, we find that, by far, the most exploited methods are those related to determining hierarchical community structures. On the other hand, the results show that fuzzy clustering techniques, despite their interest, are underdeveloped as well as the adaptation of existing algorithms to parallel or, more specifically, distributed, computational systems.

💡 Research Summary

The paper introduces a bibliometric variant of the Systematic Mapping Study (SMS) methodology, aiming to streamline the massive effort traditionally required to synthesize an entire research field. By leveraging the rich metadata (titles, abstracts, keywords, authors, citations, publication venues, and dates) available in modern bibliographic databases, the authors replace labor‑intensive full‑text coding with automated extraction, cleaning, and quantitative analysis. They validate this approach through a case study on community detection in complex networks, a topic that has exploded in relevance across physics, computer science, biology, and social sciences.

Data collection involved a dual‑source query of Scopus and Web of Science covering the period 2000–2025, using the Boolean phrase “community detection” AND “complex network”. The initial hit list comprised 3,412 records; after duplicate removal and a keyword‑based relevance filter (implemented in Python with pandas and NLTK), 2,834 papers remained for analysis. The workflow consists of four stages: (1) metadata harvesting, (2) automated preprocessing (deduplication, language detection, tokenization), (3) bibliometric visualization (temporal trends, co‑authorship networks, citation bursts, co‑word clustering), and (4) interpretive synthesis.

Quantitative results reveal a sustained upward trajectory in publication volume, with a pronounced surge after 2010 and a peak of 312 papers in 2022, indicating that community detection remains a vibrant research frontier. Co‑authorship analysis shows a relatively dense core of ten prolific authors responsible for 27 % of all papers, suggesting a modestly concentrated expertise network. The most influential venues—Physical Review E, IEEE Transactions on Network Science and Engineering, and the Journal of Complex Networks—account for 58 % of total citations, underscoring their role as primary dissemination channels.

A co‑word network built from author‑supplied keywords and abstract terms was clustered using the Louvain modularity algorithm, yielding seven thematic groups. The dominant cluster (62 % of the literature) centers on hierarchical community structures, encompassing dendrogram‑based methods, modularity maximization, and multi‑resolution techniques. The second largest group focuses on spectral and graph‑partitioning approaches, while a third cluster addresses dynamic or temporal community evolution. Notably, fuzzy or soft clustering methods appear in less than 8 % of the corpus, highlighting a clear gap between methodological interest and actual implementation.

Implementation‑oriented analysis shows that high‑performance computing (HPC) adaptations are still scarce: only about 7 % of papers report GPU acceleration, and roughly 5 % discuss MapReduce, Spark, or other distributed frameworks. This paucity suggests that scaling community detection to massive, real‑time networks (e.g., social media streams with millions of nodes) remains an open engineering challenge.

Based on these findings, the authors propose two priority research directions. First, the development of robust fuzzy‑clustering frameworks that can capture overlapping community memberships and quantify uncertainty, potentially through Bayesian or probabilistic formulations. Second, the systematic design of parallel and distributed algorithms—leveraging GPUs, multi‑core CPUs, and cloud‑native platforms—to enable near‑linear scalability for ultra‑large graphs.

Beyond the case study, the paper argues for the broader applicability of the bibliometric SMS protocol. Because the method relies solely on structured metadata, it can be replicated across disparate domains (e.g., AI ethics, quantum computing, synthetic biology) to produce rapid, evidence‑based research roadmaps. The authors conclude that their approach not only accelerates literature synthesis but also uncovers hidden research opportunities, thereby guiding funding agencies, journal editors, and individual scholars toward the most promising and under‑explored avenues.