Master’s research · Biomedical NLP
BERT-KenMeSH
Research into automatic Medical Subject Headings indexing: a multi-label classification problem for assigning relevant concepts to biomedical articles.
Research question
Can document language and label relationships work together?
The work combined PubMedBERT document representations and a dilated CNN with candidate masks from similar articles, label attention and a graph over MeSH relationships.
PubMedBERT + dilated CNNdocument features
KNN + journal signalscandidate MeSH masks
MeSH embeddings + GCNlabel relationships
Label-aware outputranked MeSH predictions
Evaluation
Report the metric with its meaning and scope.
Up to 94% Precision@1Reported evaluation result. Publish alongside the exact dataset split, experimental setting and linked thesis or paper.
What this establishes
Experience with NLP research, model development and evaluation—not professional experience operating modern LLM systems in production.
How it connects to current work
It provides a foundation for current exploration of embeddings, retrieval quality and evaluation harnesses without pretending the two eras are identical.