GHS-NET a generic hybridized shallow neural network for multi-label biomedical text classification.

摘要: Exponential growth of biomedical literature and clinical data demands more robust yet precise computational methodologies to extract useful insights from perform accurate assignment disease-specific codes. Such approaches can largely enhance the effectiveness diverse biomedicine bioinformatics applications. State-of-the-art text classification either solely leverage discrimintaive features extracted through convolution operations performed by deep convolutional neural network or contextual information recurrent network. However, none methodology takes advantage both networks. Further, existing lack produce decent performance for different genre such as notes. We, very first time, present a generic learning based hybrid multi-label namely GHS-NET which be utilized accurately classify genre. makes use most discriminative bi-directional Long Short-Term Memory acquire information. is evaluated extreme ICD-9 codes For task classification, comparison GHS-Net state-of-the-art reveals that marks increment 1%, 6%, 1% hallmarks cancer dataset, 10%, 16%, 11% chemical exposure dataset in terms precision, recall, F1-score. notes outperforms previous best over Medical Information Mart Intensive Care (MIMIC-III) significant margin 8% recall available web service at1 potentially used multi-variate disease specific text.