Past and future uses of text mining in ecology and evolution

Maxwell J. Farrell, Liam Brierley, Anna Willoughby, Andrew Yates, Nicole MideoProceedings of the Royal Society B Biological Sciences 2022 · DOI

TL;DR

Ecology and evolutionary biology are drowning in a sea of publications, making it impossible for humans to keep up. This paper argues that text mining and Natural Language Processing (NLP) are crucial tools to navigate this information overload, enabling efficient literature synthesis, data extraction, and the construction of comprehensive biodiversity databases.

What Problem Does This Paper Address?

The sheer volume of scientific literature is growing exponentially, making it incredibly difficult for researchers in ecology and evolutionary biology to conduct thorough literature reviews, meta-analyses, and build comprehensive datasets. This "synthesis gap" hinders scientific progress by making it harder to identify transformative ideas and track research trends. While text mining and NLP have revolutionized fields like biomedicine for tasks like building protein interaction databases, their adoption in ecology and evolution lags significantly, leaving a substantial opportunity for advancement.

Methods

The paper discusses various applications of text mining and NLP, broadly categorized into:

  • Document Classification: Using machine learning models to predict whether a document is relevant to a specific database or research question. This helps prioritize literature for manual review.
  • Named Entity Recognition (NER): Identifying and categorizing specific terms within texts, such as species names, locations, genes, or proteins. This is crucial for extracting structured information.
  • Relation Extraction: Identifying the relationships between these named entities. This can range from simple co-occurrence analysis to more complex parsing of sentence structures to understand how concepts are linked (e.g., species interactions, gene-disease relationships).
  • Ontology Creation and Integration: Developing structured vocabularies and knowledge graphs to harmonize and connect information across different datasets and texts, facilitating cross-disciplinary research.

The authors highlight the potential of adapting techniques from other fields, particularly biomedicine, and leveraging deep learning models for these tasks.

Datasets

The paper does not explicitly list datasets used for training or evaluation of specific NLP models. However, it references:

  • Existing literature databases: Such as the PREDICTS database (for biodiversity responses to human impacts) and various biomedical databases (e.g., for protein-protein interactions, drug-drug interactions).
  • Published articles: Both abstracts and full texts from ecology and evolutionary biology journals are considered the primary data source.
  • Grey literature and websites: Including social media posts and online texts, which can provide valuable insights into public interest and emerging trends.
  • Archival samples and natural history collections: Highlighting the potential for digitizing and applying NLP to older texts.
  • Open access repositories: Such as the PMC Open Access Subset and The General Index, which provide access to large volumes of scientific text.

Key Insights and Ideas

  • Biomedicine as a Model: The significant lag in NLP adoption in ecology and evolution compared to biomedicine presents a clear opportunity to learn from and adapt successful strategies.
  • Beyond Bibliometrics: While text mining has been used for bibliometric analysis (tracking trends), its true power lies in extracting and synthesizing actual scientific data.
  • Bridging the "Synthesis Gap": NLP tools are essential for overcoming the overwhelming volume of literature, enabling more efficient and reproducible evidence synthesis and meta-analyses.
  • Unlocking Hidden Data: Vast amounts of ecological and evolutionary knowledge are locked within unstructured text. NLP can unlock this data, leading to new discoveries and more comprehensive datasets.
  • The Power of Ontologies: Integrating ontologies is key to harmonizing diverse data and understanding complex relationships, crucial for interdisciplinary research.
  • Ethical Considerations: The paper touches upon the need for open access to scientific literature to ensure reproducibility and equitable access to information.

Key Results

The paper focuses on presenting potential and emerging applications rather than reporting specific quantitative results from novel model development. However, it highlights:

  • Improved Efficiency: Text mining can dramatically accelerate tasks like literature screening and data extraction, which are currently time-consuming and manual.
  • Enhanced Reproducibility: Automated processes improve the transparency and reproducibility of research, especially in literature reviews and database construction.
  • Identification of Research Gaps: Text analysis can systematically identify under-researched areas and emerging topics within scientific literature.
  • Successful Adaptations: Examples like the TaxoNERD model demonstrate the feasibility of adapting deep learning models for ecological tasks, even with limited domain-specific training data.

Limitations and Future Work

Limitations:

  • Limited Adoption: The primary barrier is the current low adoption rate of NLP tools in ecology and evolutionary biology.
  • Lack of Domain-Specific Tools: There's a scarcity of NLP tools specifically designed for ecological and evolutionary terminology.
  • Access to Literature: Paywalls and copyright restrictions limit access to full-text articles, hindering large-scale text mining.
  • Language Bias: The dominance of English in scientific publishing means non-English literature is often omitted.
  • Data Scarcity: The lack of readily available, domain-specific "gold standard" training data for NLP models is a significant challenge.

Future Work:

  • Promote Awareness: Increase understanding and adoption of text mining and NLP among researchers.
  • Foster Collaboration: Encourage interdisciplinary collaborations between biologists and computational linguists.
  • Develop Open-Source Tools: Create and share domain-specific NLP software and resources.
  • Build Open Access Corpora: Establish publicly accessible collections of ecological and evolutionary texts.
  • Create Expert-Created Training Data: Develop high-quality datasets for training NLP models.
  • Expand Language Support: Develop NLP tools that can process non-English scientific texts.
  • Integrate Ontologies: Further integrate ontologies with NLP software for more sophisticated data analysis.
  • Automated Alert Systems: Develop systems that can automatically alert researchers to relevant new publications and extract data from them.

Summary updated

Abstract

Ecology and evolutionary biology, like other scientific fields, are experiencing an exponential growth of academic manuscripts. As domain knowledge accumulates, scientists will need new computational approaches for identifying relevant literature to read and include in formal literature reviews and meta-analyses. Importantly, these approaches can also facilitate automated, large-scale data synthesis tasks and build structured databases from the information in the texts of primary journal articles, books, grey literature, and websites. The increasing availability of digital text, computational resources, and machine-learning based language models have led to a revolution in text analysis and natural language processing (NLP) in recent years. NLP has been widely adopted across the biomedical sciences but is rarely used in ecology and evolutionary biology. Applying computational tools from text mining and NLP will increase the efficiency of data synthesis, improve the reproducibility of literature reviews, formalize analyses of research biases and knowledge gaps, and promote data-driven discovery of patterns across ecology and evolutionary biology. Here we present recent use cases from ecology and evolution, and discuss future applications, limitations and ethical issues.
BibTeX
@inproceedings{farrell2022past,
  author = {Maxwell J. Farrell and Liam Brierley and Anna Willoughby and Andrew Yates and Nicole Mideo},
  title = {{Past and future uses of text mining in ecology and evolution}},
  booktitle = {Proceedings of the Royal Society B Biological Sciences},
  year = {2022},
  doi = {10.1098/rspb.2021.2721},
}
Figure