Temporal relation classification with deep neural network

Abstract

We proposed neural network architecture based on Convolution Neural Network(CNN) for temporal relation classification in sentence. First, we transformed word into vector by using word embedding. In Feature Extraction, we extracted two type of features. Lexical level feature considered meaning of marked entity and Sentence level feature considered context of the sentence. Window processing was used to reflect local context and Convolution and Max-pooling operation were used for global context. We concatenated both feature vectors and used softmax operation to compute confidence score. Because experiment results didn't outperform the state-of-the-art methods, we suggested some future works to do.