Research On Extraction Method Of Adverse Drug Reactions Based On Convolutional Neural Network

Adverse drug reaction(ADR)is a harmful reaction that can hurt human health during medication.Using the method to extract ADR can detect and extract ADR information from text,thus preventing medical malpractice and having great significance in practical application.However,accurate extraction of ADR is still difficult due to discontinuous keywords and complex semantics of unstructured text.To alleviate the above difficulties,two methods are proposed for the extraction of ADR.The main contents are summarized as follows:(1)A Feature Cascade Convolutional Neural Network is designed for ADR classification with the difficulty of discontinuous keywords.The feature cascade method can capture optimal feature relations between different positions of words by gradually fusing information of adjacent words and digging deep connections between multiple words to capture the keyword information.Besides,this network can mitigate the interference of imbalance in data classes by using an under-sampling method.(2)A word vector enhancement network is designed for ADR extraction entities with the difficulty of complex semantics about text.The word vector enhancement method can extract more comprehensive and accurate text semantics information by capturing the fusion semantics information of words and phrases in the vector space,alleviating the interference of complex semantics.In addition,this network can effectively mitigate the interference of irregular text expression by integrating a character embedding method with the word vector enhancement method.(3)The experiment shows that the feature cascade neural network can improve the classification performance of ADR compared with other comparative methods and effectively capture keywords information.The word vector enhancement network achieves competitive performance compared with other comparative methods and can effectively obtain text semantic information.