COVID-19: A Semantic-Based Pipeline for Recommending Biomedical Entities

Márcia Barros, Andre Lamurias, Diana F. Sousa, Pedro Ruas, Francisco M. Couto

Abstract

The rapid expansion of COVID-19 literature presents a significant challenge for researchers attempting to stay abreast of relevant biomedical entities. This paper presents a semantic-based recommendation pipeline designed to suggest specific genes, drugs, and phenotypes. By integrating multidisciplinary ontologies (GO, HP, DO, and ChEBI) with collaborative filtering on an author-entity matrix derived from the CORD-19 dataset, the system achieves a recommendation precision of 80%. Our findings demonstrate that combining diverse biological fields significantly outperforms single-ontology approaches, offering a robust tool for navigating large-scale scientific datasets.

Introduction

The COVID-19 pandemic triggered an unprecedented surge in scientific publishing, creating a unique "infodemic." For scientists, identifying relevant biomedical entities—such as specific proteins or chemical compounds—within tens of thousands of new papers is a daunting task. Traditional keyword-based search engines often fail to capture the nuanced, interconnected nature of biomedical research. This work proposes a semantic approach that suggests specific entities for investigation rather than just documents.

The Semantic Pipeline

The proposed system transforms raw text into actionable recommendations through a multi-stage process involving entity extraction, interest modeling, and collaborative filtering.

  1. Named Entity Recognition (NER) & Linking: The pipeline identifies entities in text and links them to four major ontologies: Gene Ontology (GO), Human Phenotype Ontology (HP), Disease Ontology (DO), and Chemical Entities of Biological Interest (ChEBI).
  2. Implicit Feedback Matrix: Using the LIBRETTI methodology, authorship is used as a proxy for interest. If a researcher publishes a paper mentioning an entity, it is treated as a "rating" for that entity.
  3. Collaborative Filtering (CF): The Alternating Least Squares (ALS) algorithm is applied to the author-entity matrix to recommend entities based on the publication patterns of similar researchers.
  4. Relation Extraction (RE): Relationships between entities are extracted to provide context for future explainable recommendations.
The multi-stage recommendation pipeline architecture.

Data and Evaluation

The pipeline was evaluated using the CORD-19 dataset. A subset of 9,000 full-text documents was used for the NER phase, resulting in the creation of the CORD-19-RD recommendation dataset. This dataset includes approximately 45,000 authors and 13,000 unique entities, encompassing over 3.8 million implicit ratings. To validate the automated NER module, a sample of 90 paragraphs was manually annotated by experts, yielding an F1-score of approximately 0.79.

Results

The most critical finding is the impact of multidisciplinary data on recommendation quality. When the system utilized only a single ontology (e.g., just chemicals or just diseases), precision remained low, near 20%. However, by combining all four ontologies (represented as ALS_all), the Precision@k reached 80%.

Performance comparison of the ALS algorithm across different ontology combinations, showing Precision@k, Recall@k, and MRR@k.

This performance boost highlights the value of cross-field discovery. The system can recommend entities from biological domains a researcher has not previously published in, facilitating the discovery of novel connections between virology, pharmacology, and genetics. For example, a virologist might be recommended a specific chemical compound based on the patterns of other researchers who bridge these fields.

Discussion and Future Work

While the pipeline shows high precision, challenges remain regarding entity relevance. Highly frequent but general terms (e.g., the general concept of "Disease") may be recommended often without providing specific value. Future iterations will focus on refining relevance scores and integrating Relation Extraction to provide explainable recommendations, helping researchers understand the underlying biological link behind a suggestion.

Conclusion

By leveraging semantic linking and collaborative filtering, this pipeline offers a scalable solution to information overload in biomedical research. It successfully transforms the CORD-19 "infodemic" into a structured resource for discovery, significantly improving the precision of entity-level recommendations through a multidisciplinary approach.

Summary updated

Abstract

With the increasing number of publications about COVID-19, it is a challenge to extract personalized knowledge suitable for each researcher. This work aims to build a new semantic-based pipeline for recommending biomedical entities to scientific researchers. To this end, we developed a pipeline that creates an implicit feedback matrix based on Named Entity Recognition (NER) on a corpus of documents, using multidisciplinary ontologies for recognizing and linking the entities. Our hypothesis is that by using ontologies from different fields in the NER phase, we can improve the results for state-of-the-art collaborative-filtering recommender systems applied to the dataset created. The tests performed using the COVID-19 Open Research Dataset (CORD-19) dataset show that when using four ontologies, the results for precision@k, for example, reach the 80%, whereas when using only one ontology, the results for precision@k drops to 20%, for the same users. Furthermore, the use of multi-fields entities may help in the discovery of new items, even if the researchers do not have items from that field in their set of preferences.
BibTeX
@inproceedings{barros2020covid,
  author = {Márcia Barros and Andre Lamurias and Diana F. Sousa and Pedro Ruas and Francisco M. Couto},
  title = {{COVID-19: A Semantic-Based Pipeline for Recommending Biomedical Entities}},
  booktitle = {NLP4COVID @ EMNLP},
  year = {2020},
  doi = {10.18653/v1/2020.nlpcovid19-2.20},
}
Figure