This newsletter is written automatically by an AI model (Claude) from the abstracts of the week's papers, and is not reviewed before publication. It can contain errors: follow the links to the papers before relying on a statement.
Research topics
The papers are grouped into topics at three levels of detail, chosen above. Each segment's size is its number of papers. Click a topic to select it and open its more detailed subtopics around it; click the centre to go back.
The map of the research
Each dot is a paper, placed so that papers with similar abstracts lie close together. The selected topic is highlighted; the labels mark where each topic lies. Distances between distant groups mean little, only neighbourhoods do.
How the topics grew
Papers published per year in each topic (grey lines), with the selected topic highlighted. The current year is left out because it is still incomplete.
Where in the ocean
Papers per ocean basin, from the places named in their abstracts. Darker basins have more papers (of the selected topic, or of all topics).
Topics in a basin
The topics with the most papers in the chosen basin.
Topics by ocean basin
Number of papers of each topic (rows) in each basin (columns); darker cells have more papers, on a logarithmic scale so that small counts stay visible. "No basin" holds the papers that name no ocean place, such as laboratory, freshwater or land studies.
Citation impact
Median citations per year of the papers of each topic, counting only papers at least years old. Higher means the topic's typical paper is cited more often; citation habits differ between fields, so compare with care.
How concentrated attention is
Each curve sorts a topic's papers from least to most cited and shows the share of all its citations held by the least-cited share of papers. The further a curve sags below the diagonal (perfect equality), the more the citations go to a few papers.
Citation inequality over time
The Gini coefficient of the citations per year of each topic's papers, by publication year: 0 means every paper is cited equally, 1 that one paper gets all the citations. Shown only for topic-years with at least cited papers, as fewer give unreliable values.
All topics as a table
How this review works
The review covers the scientific papers indexed in Scopus that mention marine debris, plastic pollution, microplastics, nanoplastics, microfibres or plastic pellets together with the sea, the coast, the ocean or sediments. It started from papers and grows every week.
- Collecting papers. Every Monday, the papers added to Scopus during the past week are fetched with their abstracts; duplicates and papers without an abstract are left out.
- Grouping papers into topics. Each abstract is turned into a numerical "fingerprint" of its meaning by a language model (), and papers with similar fingerprints are grouped into topics by a clustering algorithm (BERTopic with UMAP and HDBSCAN). The detailed topics are then merged, most similar first, into medium and broad topics. New papers join the topic of the most similar existing papers; the topics themselves are only rebuilt from scratch occasionally (see below).
- Naming topics. Each topic is named and described by an AI model (Claude), from its characteristic words and its most typical abstracts.
- Locating papers. Place names in each abstract are recognised automatically, looked up in the GeoNames gazetteer and assigned to the nearest ocean basin; a paper belongs to the basin most of its places fall in.
- Citations. Citation counts come from Scopus and are refreshed monthly. To compare papers of different ages, citations are divided by the paper's age, and papers younger than years are left out of citation analyses.
The topic search compares the meaning of a search with each topic's name, description and characteristic words, using a small language model () that runs in your browser, so searches are not sent anywhere. It finds topics, not individual papers or places.
How reliable are the topics?
Versions of the topics
The topics are rebuilt from scratch only when many new papers no longer fit them; each rebuild starts a new version. Between rebuilds, papers are added to the existing topics and past assignments never change.
Limitations
- Coverage. Only papers indexed by Scopus and matching the search are included; books, reports and papers in other databases are not.
- Locations. Basins are inferred from the place names in abstracts, which can be missing, ambiguous (a city name shared by two countries) or about where the authors work rather than where they sampled. of papers name no ocean place at all.
- Topics. Topics are statistical groupings of similar abstracts, not an expert classification. A paper belongs to a single topic even when it touches several, and about of papers fit no topic well and are left out of the topic charts.
- Citations. Citation counts favour older papers and fields with more researchers; the median and Gini coefficient describe each topic's citations, not the quality of its research.
- AI-written text. Topic names, descriptions and the weekly newsletter are written by an AI model () and published without human review. They can be inaccurate.
Data and code
The code and data of this review are open: github.com/henrique-back/mpl_topics_live. It continues the analysis of an earlier study (preprint on SSRN; code at github.com/henrique-back/mpl_topics).
References
- Scopus, the source of papers, abstracts and citation counts. scopus.com
- Qwen3-Embedding, the language model that turns abstracts into fingerprints. huggingface.co/Qwen/Qwen3-Embedding-4B
- BERTopic, the topic-modelling framework. maartengr.github.io/BERTopic
- UMAP, which reduces the fingerprints to a few dimensions before clustering and draws the map of the research. umap-learn.readthedocs.io
- HDBSCAN, the clustering algorithm that finds the topics. hdbscan.readthedocs.io
- Consensus clustering. The topics combine many runs of UMAP and HDBSCAN by evidence accumulation: Fred, A. L. N. and Jain, A. K. (2005). Combining multiple clusterings using evidence accumulation. IEEE Transactions on Pattern Analysis and Machine Intelligence 27(6), 835–850. doi.org/10.1109/TPAMI.2005.113
- spaCy, which recognises place names in the abstracts. spacy.io
- GeoNames, the gazetteer that gives each place its coordinates. geonames.org
About
The Marine Plastics Research Monitor is made and maintained by Henrique Back. It grew out of a topic-modelling study of the microplastics literature and keeps that analysis running as new papers come out, so that researchers, policy makers and the public can see what is being studied, where, and how much.
The work was developed in partnership with the Laboratório de Engenharia e Ciências Oceânicas (LabECO) of the Universidade Federal de Santa Catarina (UFSC), Brazil, and the MERS research group of the Institute of Environmental Science and Technology of the Universitat Autònoma de Barcelona (ICTA-UAB), Spain.
The study
The method and the first results are described in a preprint by Henrique de Medeiros Back, Michael Grelaud, Orestes Estevam Alarcon and Patrizia Ziveri. Please cite it when you use this review:
Back, H. de M., Grelaud, M., Alarcon, O. E. and Ziveri, P. Decoding the Evolution of Marine Plastic Litter Research with Large Language Models. Preprint, SSRN. doi.org/10.2139/ssrn.5332715
Contact
Questions, corrections or ideas are welcome, including reports of a wrongly named topic or an error in the newsletter.
- Email: henriback2@gmail.com
- LinkedIn: linkedin.com/in/henriback
- GitHub: github.com/henrique-back