Team Fusion@ SU@ BC8 SympTEMIST track: transformer-based approach for symptom recognition and linking

Georgi Grazhdanski; Sylvia Vassileva; Ivan Koychev; Svetla Boytcheva

cs.CLApr 7, 2026

Team Fusion@ SU@ BC8 SympTEMIST track: transformer-based approach for symptom recognition and linking

Authors: Georgi Grazhdanski, Sylvia Vassileva, Ivan Koychev, Svetla Boytcheva

Organizations: FMI, Sofia University St. Kliment Ohridski · Ontotext

Abstract

This paper presents a transformer-based approach to solving the SympTEMIST named entity recognition (NER) and entity linking (EL) tasks. For NER, we fine-tune a RoBERTa-based (1) token-level classifier with BiLSTM and CRF layers on an augmented train set. Entity linking is performed by generating candidates using the cross-lingual SapBERT XLMR-Large (2), and calculating cosine similarity against a knowledge base. The choice of knowledge base proves to have the highest impact on model accuracy.

Figures & tables

Model	Val P	Val R	Val F1	Test P	Test R	Test F1
Aug. Spanish RoBERTa + CRF	0.773	0.729	0.750	0.732	0.718	0.725
Aug. Spanish RoBERTa + BiLSTM + CRF	0.744	0.732	0.738	0.739	0.725	0.732
Pre-trained Aug. Spanish RoBERTa + CRF	0.749	0.721	0.735	0.715	0.720	0.718
CLIN-X-ES + CRF	0.757	0.717	0.737	0.718	0.703	0.710
Aug. CLIN-X-ES + CRF	0.722	0.704	0.713	0.724	0.699	0.712

Table 1: Subtask 1 results on the validation and test sets.

Model	Val P	Val R	Val F1	Test P	Test R	Test F1
Spanish RoBERTa + CRF	0.730	0.746	0.738	-	-	-
Augmented Spanish RoBERTa + CRF	0.773	0.729	0.750	0.732	0.718	0.720
CLIN-X-ES + CRF	0.757	0.718	0.737	0.718	0.703	0.710
Augmented CLIN-X-ES + CRF	0.722	0.704	0.713	0.724	0.699	0.712

Table 2: Effect of data augmentation in the NER subtask.

Model	Knowledge Base	Val	Test
		Acc	Acc
SapBERT XLMR-Large	Gazetteer + Train	0.514	0.588
SapBERT XLMR-Large	Gazetteer + Train + Aug.	0.533	0.565
SapBERT XLMR-Large	Gazetteer + Train + UMLS	0.524	0.589
SapBERT XLMR-Large + Sliding Window	Gazetteer + Train + Aug.	0.536	0.587
SapBERT XLMR-Large	Gazetteer + Train + UMLS	0.510	0.017

Table 3: Subtask 2 results on the validation and test sets.

Named Entity Recognition

Named Entity Recognition identifies and classifies people, places, organizations, and other entities in text, speech, and multimodal data.

See all 106 papers

BELXTR: Biomedical Entity Linking via Contextualized Token Retrieval
Sep 22, 2026Samuele Garda, Ulf LeserLate Interaction Retrieval Biomedical Entity Linking
Computer Science, Humboldt-Universität zu Berlin, Rudower Chaussee 25, 12489, Berlin, Germany
Learning Proposition-Level Links from Sentence-Level Supervision for Neuro-Symbolic NLI
Oct 10, 2026Xuyao Feng, Antonis BikakisNatural Language Inference Neuro-Symbolic Reasoning
University College London, United Kingdom
Named Entity Recognition using Sliding Window Approach
Aug 30, 2026Hariom Ingle, Ronit Ghode, Ishwari Gondkar +3Named Entity Recognition
Department of Information Technology, PICT, Pune, India · L3Cube Labs, Pune, India · Indian Institute of Technology Madras, Chennai, India

Sep 22, 2026J/K move · Enter open · S save

BELXTR: Biomedical Entity Linking via Contextualized Token Retrieval

Samuele Garda, Ulf Leser

Computer Science, Humboldt-Universität zu Berlin, Rudower Chaussee 25, 12489, Berlin, Germany

Late Interaction Retrieval Biomedical Entity Linking Multi-Vector Retrieval Entity Linking

Open paper AskPDF arXiv

Biomedical Entity Linking disambiguates mentions to entities in a knowledge base (KB), making it the cornerstone of information extraction pipelines. While embedding-based models are a popular approach for the task, they suffer from a key limitation. They compress mentions (and entities) into a single vector, forcing the model to average away crucial fine-grained differences. We present BELXTR, a novel embedding model based on the multi-vector (a.k.a. late interaction) architecture, which allows to leverage token-level matching information. BELXTR extends the original XTR model to biomedical entity linking by integrating an existing task-specific training objective and exploring active query expansion. Experiments across ten corpora and five KBs show that BELXTR improves upon current state-of-the-art in half of the corpora with an average improvement of 5pp recall@1. The largest gains are reported on the challenging cross-species gene disambiguation subtask, where BELXTR outperforms an LLM-powered retrieve-and-rerank pipeline and closely approaches a specialized rule-based system. Our results highlight multi-vector models as a practical alternative to hard-to-maintain rule-based systems or in scenarios where LLM-based reranking is too costly as in PubMed-scale mining. The code to reproduce our experiments can be found at: https://github.com/sg-wbi/belxtr.

Abstract

Named Entity Recognition

Explore similar work

BELXTR: Biomedical Entity Linking via Contextualized Token Retrieval

Learning Proposition-Level Links from Sentence-Level Supervision for Neuro-Symbolic NLI

Named Entity Recognition using Sliding Window Approach