KFU at CLEF eHealth 2017 Task 1: ICD-10 coding of English death certificates with recurrent neural networks

Abstract

This paper describes the participation of the KFU team in the CLEF eHealth 2017 challenge. Specifically, we participated in Task 1, namely "Multilingual Information Extraction - ICD-10 coding" for which we implemented recurrent neural networks to automatically assign ICD-10 codes to fragments of death certificates written in English. Our system uses Long Short-Term Memory (LSTM) to map the input sequence into a vector representation, and then another LSTM to decode the target sequence from the vector. We initialize the input representations with word embeddings trained on user posts in social media. The encoderdecoder model obtained F-measure of 85.01% on a full test set with significant improvement as compared to the average score of 62.2% for all participants' approaches. We also obtained significant improvement from 26.1% to 44.33% on an external test set as compared to the average score of the submitted runs.