Research And Platform Development On Chinese Electronic Medical Records Entity Assertion And Relation Extraction Based On Deep Learning

With the improvement of people’s life quality and the development of economic level, medical and health problems have been widely concerned by the society. And as nowadays electronic medical system rapidly grows popular, a large number of medical related data was stored by EMR. Through continuous accumulation, various forms of electronic medical system produced huge medical data. The Chinese EMR data has had a preliminary scale. Traditional natural language processing technology mostly require manual annotation of these data, which leads the original data that unlabeled cannot be effectively applied to task processing. By using the characteristic of deep learning algorithm, the futures by self-learning can be used to train unlabeled data well with unsupervised learning. At the same time, the words embedding in the training process itself contains the context information, which can be used for the input of neural network model to achieve better results.In this paper, needle for the characteristics of Chinese EMR, developed an automated, scalable and highly modular deep learning algorithm platform, integrated recurrent neural network and recursive neural tensor networks model, which is popular in the fields of deep learning, for processing common tasks on Chinese EMR.Based on the development of deep learning platform, this paper also makes a research and experiment on the Chinese EMR entity relation extraction and entity assertion, and obtains some experimental results. Entity relation extraction using recursive neural tensor networks, in constructing the semantic tree training word vector recursive, and training multiple classifiers by top-down achieve classification relationship.Entity assertion recognition using recurrent neural network, mainly set the semantic information around the entity as the features to train classification of entity assertion.