Bibliographic Exploration Methodologies to Support Systematic and Scoping Reviews
This paper proposes and demonstrates a replicable framework leveraging Natural Language Processing and machine learning techniques to analyze bibliographic corpora, thereby assisting researchers in formulating research questions, identifying thematic gaps, and refining search strategies prior to conducting formal systematic or scoping reviews.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine trying to understand a vast, ever-growing library where new books are added every second, written in different languages, and organized by no single system. This is the reality of modern science. Researchers today face a flood of information so immense that simply finding the right books to read for a specific project has become a monumental task. When scientists want to understand a topic, they often conduct a "scoping review," a careful process of mapping out what is already known, identifying gaps, and organizing the chaos. But before they can even start reading the full texts, they must first gather thousands of potential articles from massive databases. The problem is that these initial lists are often too large and messy to handle by hand. A new study suggests that researchers can use computer tools to get a clear, high-level view of this mountain of literature before they begin the slow work of reading every single page.
The researchers, a team from the Federal University of ABC in Brazil, set out to test a method that uses artificial intelligence to explore scientific literature. They focused on a specific, critical field: how Brazil manages its municipal solid waste. In January 2026, they searched two of the world's largest scientific databases, Scopus and Web of Science, using keywords related to solid waste and Brazil. Their search returned 1,336 records. After removing duplicates, they were left with 1,281 academic works, including journal articles, book chapters, and conference papers. Instead of reading every title and abstract one by one, they fed this entire collection into a computer program designed to find patterns in language.
The team treated the titles and abstracts of these papers as a single, massive text. They first cleaned the data, removing common words like "the" or "and" that do not carry much meaning, and focused on the significant terms. They then used several different computer techniques to see how the words and ideas connected. One method looked at which words frequently appeared together, creating a map of relationships. Another method grouped the documents into clusters based on how similar their vocabulary was, much like sorting a pile of mixed-up cards into neat decks without knowing the rules beforehand. They also used two advanced statistical approaches to uncover hidden themes, or "topics," that ran through the collection. One of these approaches, which they found to be the most effective for this specific set of texts, broke the documents down into distinct groups based on the combination of words they contained.
The results revealed a clear structure within the literature. The computer analysis showed that the research on solid waste in Brazil is not a scattered collection of random ideas but is organized around a central core. This core connects the physical act of managing waste with four major dimensions: environmental concerns, institutional policies, operational logistics, and socioeconomic factors. The computer maps showed that terms like "landfill," "recycling," and "emissions" formed one tight group, while words like "policy," "public service," and "municipal" formed another. A third group linked the economic aspects, such as "cost" and "value," with social issues like "health" and "population." By visualizing these connections, the researchers could see that the field is dominated by the practical challenges of waste management in the Brazilian context, with clear sub-areas emerging around specific processes like reverse logistics and biogas energy.
Crucially, the study did not find that the computer could replace the human researcher. The authors were careful to state that their goal was not to automate the review process or to let a machine decide which papers are important. Instead, the computer served as a powerful lens that helped the researchers see the shape of the field before they started their detailed work. The analysis suggested that the most effective way to organize this specific type of literature was through a method that creates distinct, non-overlapping groups of topics, rather than one that allows themes to blend together too much. This distinction helped the team understand that while some research areas are very specific and separate, others naturally overlap, and the computer tools could highlight where those boundaries lie.
The team concluded that this approach offers a replicable way to prepare for formal reviews. By using these tools, researchers can formulate better questions, refine their search strategies, and identify where the gaps in knowledge truly are before they spend months reading individual papers. The study demonstrates that when faced with a deluge of information, computational methods can provide a structured, transparent, and evidence-based way to take the first step. It allows scientists to move from being overwhelmed by the sheer volume of data to having a clear, panoramic view of the landscape, ensuring that when they do begin their deep dive, they are looking in the right direction.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.