Medical Knowledge-enriched Textual Entailment Framework

Abstract

One of the cardinal tasks in achieving robust medical question answering systems is textual entailment. The existing approaches make use of an ensemble of pre-trained language models or data augmentation, often to clock higher numbers on the validation metrics. However, two major shortcomings impede higher success in identifying entailment: (1) understanding the focus/intent of the question and (2) ability to utilize the real-world background knowledge to capture the context beyond the sentence. In this paper, we present a novel Medical Knowledge-Enriched Textual Entailment framework that allows the model to acquire a semantic and global representation of the input medical text with the help of a relevant domain-specific knowledge graph. We evaluate our framework on the benchmark MEDIQA-RQE dataset and manifest that the use of knowledgeenriched dual-encoding mechanism help in achieving an absolute improvement of 8.27% over SOTA language models. We have made the source code available here. 1

Topical Classification of Food Safety Publications with a Knowledge Base

“…-Anatomical Entity Ontology (AEO) 12 -Agronomy Ontology (AGRO) 13 -Apollo Structured Vocabulary (APOLLO-SV) 14 -Basic Formal Ontology (BFO) 15 -BRENDA tissue / enzyme source (BTO) 16 10 https://bioportal.bioontology.org/ontologies/FOODON/?p=mappings 11 http://obofoundry.org/ 12 http://www.obofoundry.org/ontology/aeo.html 13 https://github.com/AgriculturalSemantics/agro 14 https://github.com/ApolloDev/apollo-sv 15 http://basic-formal-ontology.org/ 16 http://www.obofoundry.org/ontology/bto.html -Common Anatomy Reference Ontology (CARO) 17 -Chemical Entities of Biological Interest (CHEBI) 18 -Chemical Methods Ontology (CHMO) 19 -Cell Ontology (CL) 20 -Human Disease Ontology (DOID) 21 -Drug Ontology (DRON) 22 -Human developmental anatomy ontology (EHDAA2) 23 -Environment Ontology (ENVO) 24 -Food-Biomarker Ontology (FOBI) 25 -FoodOn 26 -Gazetteer (GAZ) 27 -Gene Ontology (GO) 28 -Human Phenotype Ontology (HP) 29 -Information Artifact Ontology (IAO) This relatively low coverage contrasts with the very wide range of topics these ontologies cover. This is possibly due to MeSH being designed for indexing publications, and thus having terms oriented toward more general topics instead of granular concepts.…”

“…A candidate entity does not provide context information by itself -it is necessary to look at related entities to obtain a better sense of what it describes. This "graph expansion" approach was used successfully in a textual entailment system [29]. Before attempting to design an algorithm for this problem, KB was examined and several observations were made.…”

Pre-trained Language Models in Biomedical Domain: A Systematic Survey

“…Sharma et al [212] incorporated the embedding of knowledge graph (UMLS) in the biomedical domain into the BioELMo to improve its performance. Yadav et al [274] a novel framework Sem-KGN for the medical textual entailment task, which infused the medical entity information from the medical knowledge bases into the BERT model. He et al [78] proposed to infuse the domain knowledge of disease into a series of pre-trained language models including BERT, BioBERT, SciBERT, ClinicalBERT, BlueBERT, and ALBERT, to improve their performance in the question answering, medical inference, and MedNLI [203] Clinical notes Entailment, contradiction, or neutral 14,049 MEDIQA-RQE [1] Consumer health questions Entailment, contradiction 9,120 CMFAQ [302] Consumer health questions Entailment, contradiction 53,822 disease name recognition task.…”