A convolutional route to abbreviation disambiguation in clinical text

Venkata Joopudi, Bharath Dandala, Murthy Devarakonda

    Research output: Contribution to journalArticle

    2 Citations (Scopus)

    Abstract

    Objective: Abbreviations sense disambiguation is a special case of word sense disambiguation. Machine learning methods based on neural networks showed promising results for word sense disambiguation (Festag and Spreckelsen, 2017) [1] and, here we assess their effectiveness for abbreviation sense disambiguation. Methods: Convolutional Neural Network (CNN) models were trained, one for each abbreviation, to disambiguate abbreviation senses. A reverse substitution (of long forms with short forms) method from a previous study was used on clinical narratives from Cleveland Clinic, USA, to auto-generate training data. Accuracy of the CNN and traditional Support Vector Machine (SVM) models were studied using: (a) 5-fold cross validation on the auto-generated training data; (b) a manually created, set-aside gold standard; and (c) 10-fold cross validation on a publicly available dataset from a previous study. Results: CNN improved accuracy by 1–4 percentage points on all the three datasets compared to SVM, and the improvement was the most for the set-aside dataset. The improvement was statistically significant at p < 0.05 on the auto-generated dataset. We found that for some common abbreviations, sense distributions mismatch between the test and auto generated training data, and mitigating the mismatch significantly improved the model accuracy. Conclusion: The neural network models work well in disambiguating abbreviations in clinical narratives, and they are robust across datasets. This avoids feature-engineering for each dataset. Coupled with an enhanced auto-training data generation, neural networks can simplify development of a practical abbreviation disambiguation system.

    Original languageEnglish (US)
    Pages (from-to)71-78
    Number of pages8
    JournalJournal of Biomedical Informatics
    Volume86
    DOIs
    StatePublished - Oct 1 2018

    Fingerprint

    Neural networks
    Neural Networks (Computer)
    Support vector machines
    Learning systems
    Datasets
    Substitution reactions
    Support Vector Machine

    ASJC Scopus subject areas

    • Computer Science Applications
    • Health Informatics

    Cite this

    A convolutional route to abbreviation disambiguation in clinical text. / Joopudi, Venkata; Dandala, Bharath; Devarakonda, Murthy.

    In: Journal of Biomedical Informatics, Vol. 86, 01.10.2018, p. 71-78.

    Research output: Contribution to journalArticle

    @article{407c1c80702546eeb48bf1f620184a75,
    title = "A convolutional route to abbreviation disambiguation in clinical text",
    abstract = "Objective: Abbreviations sense disambiguation is a special case of word sense disambiguation. Machine learning methods based on neural networks showed promising results for word sense disambiguation (Festag and Spreckelsen, 2017) [1] and, here we assess their effectiveness for abbreviation sense disambiguation. Methods: Convolutional Neural Network (CNN) models were trained, one for each abbreviation, to disambiguate abbreviation senses. A reverse substitution (of long forms with short forms) method from a previous study was used on clinical narratives from Cleveland Clinic, USA, to auto-generate training data. Accuracy of the CNN and traditional Support Vector Machine (SVM) models were studied using: (a) 5-fold cross validation on the auto-generated training data; (b) a manually created, set-aside gold standard; and (c) 10-fold cross validation on a publicly available dataset from a previous study. Results: CNN improved accuracy by 1–4 percentage points on all the three datasets compared to SVM, and the improvement was the most for the set-aside dataset. The improvement was statistically significant at p < 0.05 on the auto-generated dataset. We found that for some common abbreviations, sense distributions mismatch between the test and auto generated training data, and mitigating the mismatch significantly improved the model accuracy. Conclusion: The neural network models work well in disambiguating abbreviations in clinical narratives, and they are robust across datasets. This avoids feature-engineering for each dataset. Coupled with an enhanced auto-training data generation, neural networks can simplify development of a practical abbreviation disambiguation system.",
    author = "Venkata Joopudi and Bharath Dandala and Murthy Devarakonda",
    year = "2018",
    month = "10",
    day = "1",
    doi = "10.1016/j.jbi.2018.07.025",
    language = "English (US)",
    volume = "86",
    pages = "71--78",
    journal = "Journal of Biomedical Informatics",
    issn = "1532-0464",
    publisher = "Academic Press Inc.",

    }

    TY - JOUR

    T1 - A convolutional route to abbreviation disambiguation in clinical text

    AU - Joopudi, Venkata

    AU - Dandala, Bharath

    AU - Devarakonda, Murthy

    PY - 2018/10/1

    Y1 - 2018/10/1

    N2 - Objective: Abbreviations sense disambiguation is a special case of word sense disambiguation. Machine learning methods based on neural networks showed promising results for word sense disambiguation (Festag and Spreckelsen, 2017) [1] and, here we assess their effectiveness for abbreviation sense disambiguation. Methods: Convolutional Neural Network (CNN) models were trained, one for each abbreviation, to disambiguate abbreviation senses. A reverse substitution (of long forms with short forms) method from a previous study was used on clinical narratives from Cleveland Clinic, USA, to auto-generate training data. Accuracy of the CNN and traditional Support Vector Machine (SVM) models were studied using: (a) 5-fold cross validation on the auto-generated training data; (b) a manually created, set-aside gold standard; and (c) 10-fold cross validation on a publicly available dataset from a previous study. Results: CNN improved accuracy by 1–4 percentage points on all the three datasets compared to SVM, and the improvement was the most for the set-aside dataset. The improvement was statistically significant at p < 0.05 on the auto-generated dataset. We found that for some common abbreviations, sense distributions mismatch between the test and auto generated training data, and mitigating the mismatch significantly improved the model accuracy. Conclusion: The neural network models work well in disambiguating abbreviations in clinical narratives, and they are robust across datasets. This avoids feature-engineering for each dataset. Coupled with an enhanced auto-training data generation, neural networks can simplify development of a practical abbreviation disambiguation system.

    AB - Objective: Abbreviations sense disambiguation is a special case of word sense disambiguation. Machine learning methods based on neural networks showed promising results for word sense disambiguation (Festag and Spreckelsen, 2017) [1] and, here we assess their effectiveness for abbreviation sense disambiguation. Methods: Convolutional Neural Network (CNN) models were trained, one for each abbreviation, to disambiguate abbreviation senses. A reverse substitution (of long forms with short forms) method from a previous study was used on clinical narratives from Cleveland Clinic, USA, to auto-generate training data. Accuracy of the CNN and traditional Support Vector Machine (SVM) models were studied using: (a) 5-fold cross validation on the auto-generated training data; (b) a manually created, set-aside gold standard; and (c) 10-fold cross validation on a publicly available dataset from a previous study. Results: CNN improved accuracy by 1–4 percentage points on all the three datasets compared to SVM, and the improvement was the most for the set-aside dataset. The improvement was statistically significant at p < 0.05 on the auto-generated dataset. We found that for some common abbreviations, sense distributions mismatch between the test and auto generated training data, and mitigating the mismatch significantly improved the model accuracy. Conclusion: The neural network models work well in disambiguating abbreviations in clinical narratives, and they are robust across datasets. This avoids feature-engineering for each dataset. Coupled with an enhanced auto-training data generation, neural networks can simplify development of a practical abbreviation disambiguation system.

    UR - http://www.scopus.com/inward/record.url?scp=85052861729&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=85052861729&partnerID=8YFLogxK

    U2 - 10.1016/j.jbi.2018.07.025

    DO - 10.1016/j.jbi.2018.07.025

    M3 - Article

    C2 - 30118854

    AN - SCOPUS:85052861729

    VL - 86

    SP - 71

    EP - 78

    JO - Journal of Biomedical Informatics

    JF - Journal of Biomedical Informatics

    SN - 1532-0464

    ER -