Dual Architecture for Name Entity Extraction and Relation Extraction with Applications in Medical Corpora

Anonymous

Abstract: There is a growing interest in automatic knowledge discovery in plain text documents. Automation enables the analysis of massive collections of information. Such efforts are especially relevant in the health domain as advancements could use the large volume of available resources to transform areas important for society when addressing various health research challenges. However, knowledge discovery is usually aided by annotated corpora, which are scarce resources in the literature. This situation is particularly critical in the Spanish language, for which the volume of training resources is less widespread. This work uses a health-oriented Spanish dataset, and it also creates an English variant using the same tagging system. Furthermore, we design and analyze two separated architectures for Entity Extraction and Relation Recognition that outperform previous works in the Spanish dataset. With such promising results, we also evaluate their performance in the English version. Finally, we perform a use case experiment to evaluate the utility of the output of these two architectures in Information Retrieval systems.