Combination of Deep Recurrent Neural Networks and Conditional Random Fields for Extracting Adverse Drug Reactions from User Reviews.

Elena Tutubalina,Sergey Nikolenko

doi:10.1155/2017/9451342

Abstract

Adverse drug reactions (ADRs) are an essential part of the analysis of drug use, measuring drug use benefits, and making policy decisions. Traditional channels for identifying ADRs are reliable but very slow and only produce a small amount of data. Text reviews, either on specialized web sites or in general-purpose social networks, may lead to a data source of unprecedented size, but identifying ADRs in free-form text is a challenging natural language processing problem. In this work, we propose a novel model for this problem, uniting recurrent neural architectures and conditional random fields. We evaluate our model with a comprehensive experimental study, showing improvements over state-of-the-art methods of ADR extraction.

Highlights

Recent studies on text mining applications increasingly employ nonstandard sources of information to obtain new data related to health conditions, the efficiency of treatment, drug reactions, and interactions between different drugs
Our results show that the joint model of recurrent neural networks (RNN) and conditional random fields (CRF) improves the performance of state-of-the-art CRF and RNN models trained separately
We can summarize the contributions of this work as follows: (i) we have introduced a joint model that combines CRF and RNN to model the sequence of labels for adverse drug reactions (ADRs) extraction; (ii) we have conducted empirical evaluation of this model on benchmark datasets; and (iii) experimental results have shown that the proposed model improves over state-of-the-art performance

Summary

Introduction

Recent studies on text mining applications increasingly employ nonstandard sources of information to obtain new data related to health conditions, the efficiency of treatment, drug reactions, and interactions between different drugs. Users provide information about themselves through social media posts and free-text forum comments. This rich source of information has been successfully used, for instance, to monitor adverse drug reactions, making it possible to detect rare and underestimated reactions through the users complaining about their health [1]. A sentence “1st pill taken with food, a few hours after I experienced shortness of breath, a sense of depression, cramping, upset stomach” contains four ADRs, namely, shortness of breath, depression, cramping, and upset stomach This challenging task is divided into the two subtasks: identification of ADRs and normalization of ADRs. In this paper, we focus on the first subtask

Objectives

Results

Discussion

Conclusion