Extracting Adverse Drug Reactions using Deep Learning-and Dictionary-Based Approaches. Tiftikci, M.; Özgür, A.; He, Y.; and Hur, J. Technical Report
Extracting Adverse Drug Reactions using Deep Learning-and Dictionary-Based Approaches [pdf]Paper  Extracting Adverse Drug Reactions using Deep Learning-and Dictionary-Based Approaches [link]Website  abstract   bibtex   
Drug labels contain detailed information about the drugs including their safety concerns which are regulated by the United States Food and Drug Administration (FDA). Adverse drug reactions (ADR) are adverse reactions associated with a specific drug. Automatic extraction of ADRs could help FDA greatly regulate drug safety. In this study, we employed an integrated approach of machine learning (ML)-based and dictionary-/rule-based methods to recognize ADR terms and normalize these terms to MedDRA Preferred Terms. The machine learning approach was used for the identification of the entities and is based on a recently proposed deep learning architecture. The model includes bi-directional Long Short-Term Memory (Bi-LSTM), a Convolutional Neural Network (CNN), and Conditional Random Fields (CRF). Alternatively, a dictionary-and rule-based approach was also used to identify ADR terms. MedDRA terms were added as a dictionary to SciMiner, our in-house text-mining system, and multiple rules for term expansion and exclusion to increase coverage and accuracy were implemented. The best performance was achieved using a combined approach: ADRs were first identified by the ML-based approach and then normalized to MedDRA Preferred Terms by the dictionary-and rule-based approach. Our system achieved 76.97% F1 score on the entity detection task and 82.58% micro-averaged F1 score on the ADR normalization task in the TAC 2017 ADR challenge.
@techreport{
 title = {Extracting Adverse Drug Reactions using Deep Learning-and Dictionary-Based Approaches},
 type = {techreport},
 websites = {http://www.nltk.org},
 id = {0f1e1acf-a7c3-3f81-bc64-0176aec8fb9b},
 created = {2019-10-11T20:27:00.908Z},
 accessed = {2019-10-11},
 file_attached = {true},
 profile_id = {1971c810-6732-3a00-9f6b-d217e1a53071},
 group_id = {cbcfbfec-195f-3b99-b6a1-d26e1dd80ff5},
 last_modified = {2019-10-12T09:02:52.792Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 citation_key = {Tiftikci},
 private_publication = {false},
 abstract = {Drug labels contain detailed information about the drugs including their safety concerns which are regulated by the United States Food and Drug Administration (FDA). Adverse drug reactions (ADR) are adverse reactions associated with a specific drug. Automatic extraction of ADRs could help FDA greatly regulate drug safety. In this study, we employed an integrated approach of machine learning (ML)-based and dictionary-/rule-based methods to recognize ADR terms and normalize these terms to MedDRA Preferred Terms. The machine learning approach was used for the identification of the entities and is based on a recently proposed deep learning architecture. The model includes bi-directional Long Short-Term Memory (Bi-LSTM), a Convolutional Neural Network (CNN), and Conditional Random Fields (CRF). Alternatively, a dictionary-and rule-based approach was also used to identify ADR terms. MedDRA terms were added as a dictionary to SciMiner, our in-house text-mining system, and multiple rules for term expansion and exclusion to increase coverage and accuracy were implemented. The best performance was achieved using a combined approach: ADRs were first identified by the ML-based approach and then normalized to MedDRA Preferred Terms by the dictionary-and rule-based approach. Our system achieved 76.97% F1 score on the entity detection task and 82.58% micro-averaged F1 score on the ADR normalization task in the TAC 2017 ADR challenge.},
 bibtype = {techreport},
 author = {Tiftikci, Mert and Özgür, Arzucan and He, Yongqun and Hur, Junguk}
}
Downloads: 0