Natural language processing models balance speed and accuracy for health texts
Custom Named Entity Recognition and Topic Classification for Global Health Publications
Artificial IntelligenceComputation and Language
Summary
Identifying important terms and topics in global health articles is tough when you have little labeled data and limited computing power. The authors compared different language models, finding that bigger datasets don’t always help find better health-specific words. They tested fast convolutional models against slower but more accurate transformer models for finding health entities. For sorting articles into topics, a zero-shot transformer worked better but was slower than a smaller alternative. Their work shows how different tools can be chosen for practical needs, balancing quality and resource use.
What this means in practice
- •For hospital data teams: Use the entity recognition and topic classification pipeline to index global health documents quickly for better research retrieval and analysis.
- •For legal document analysts: Adapt the lightweight convolutional models combined with domain-specific filters to extract entities and classify topics in legal texts under resource constraints.
Authors
Genis Skura, Antoine Geissbühler, Jean-Luc Falcone
Abstract
How should natural language processing models be selected and adapted for global health literature in environments where annotated data and computational resources are limited? This thesis investigates these challenges through experiments on semantic tag discovery, named entity recognition (NER), and multi-label topic classification. First, skip-gram word2vec models trained on progressively larger specialized corpora are compared with BioWordVec to assess how corpus size and domain context influence tag discovery. Vocabulary coverage and qualitative evaluation indicate that broader coverage does not necessarily yield more useful domain-specific associations. The analysis then turns to entity extraction, comparing convolutional spaCy models with a RoBERTa-based transformer on 1,000 annotated sentences. Under a lenient scoring protocol, the transformer achieves 0.80 micro-F1 versus 0.65-0.69 for convolutional models, but takes 82 seconds rather than 5-6 seconds. This trade-off motivates fine-tuning convolutional models and integrating a disease recognizer that achieves 81.33% test F1 on the NCBI Disease Corpus. Combined with PDF preprocessing, entity filtering, and MeSH enrichment, the resulting pipeline supports document-level indexing. To complement entity extraction with thematic annotation, MiniLM-based few-shot classification is compared with BART-MNLI zero-shot inference across 50 topics and 1,000 handcrafted test sentences. BART-MNLI achieves 95.2% single-label accuracy versus 59%; reported multi-label accuracies are 88% and 32% under partly manual assessment. However, its higher inference cost limits practical integration. The results show where domain specialization and lightweight adaptation offer practical value, and where transformer accuracy justifies higher inference costs, providing an empirical basis for building knowledge systems under resource constraints.