Discriminative Neural Sentence Modeling by Tree-Based Convolution

摘要: This paper proposes a tree-based convolutional neural network (TBCNN) for discriminative sentence modeling. Our models leverage either constituency trees or dependency of sentences. The convolution process extracts sentences' structural features, and these features are aggregated by max pooling. Such architecture allows short propagation paths between the output layer underlying feature detectors, which enables effective learning extraction. We evaluate our on two tasks: sentiment analysis question classification. In both experiments, TBCNN outperforms previous state-of-the-art results, including existing networks dedicated feature/rule engineering. also make efforts to visualize process, shedding light how work.