Research On Clinic Event Recognition Based Bi-LSTM-CRF Model

Recent years,the depth of the text semantic analysis has become a hot spot in the field of natural language processing research.The entity recognition is an important part of text semantic analysis.Information extraction technology research is conducive to the further development of other text processing technology,such as information retrieval,entity recognition,text summary and so on.Deep learning technology made a great success in natural language processing areas,many scholars applied deep learning technology to sequence annotation,entity recognition tasks such as above,and on the standard test set to get the better results than traditional statistical method.Howerer,previous work still exist problems such as inadequate access to information.We propose a deep learning model utilized by bi-LSTM as well as CRF based in previous work.To finish the task of clinic event extraction and attribute prediction from 2016 SemEval,the model employs the context words together with their part-of-speech tags and named entities to compose features,then utilize the bidirectional LSTM neural network to learn the hidden feature representations.In prediction step,we employ the hidden features as the input of CRF to predict clinic event spans and its attributes.The empirical evaluation demonstrates that our approach significantly outperforms baseline methods.