This paper proposes a new convolutional neural architecture based on tree-structures, called the tree-based convolutional neural network (TBCNN). Two variants take advantage of constituency trees and dependency trees, respectively, to model sentences. Compared with traditional "flat" convolutional neural networks (CNNs), TBCNNs explore explicitly the structural information of sentences; compared with recursive neural networks, TBCNNs have much shorter propagation paths, enabling more effective feature learning and extraction. In the experiment of sentiment analysis, our two models consistently outperform CNNs and RNNs in a controlled setting. Our models are also competitive to most state-of-the-art results, including RNNs based on long short term memory and deep CNNs/RNNs.

本研究提出了一种基于树的卷积神经网络（TBCNN），用于区分性句子建模。文章通过 constituency trees 或 dependency trees 提取句子的结构特征，并通过最大池聚合这些特征。我们评估了我们的模型在情感分析和问题分类两个领域的表现，TBCNN的表现均优于以前的最新结果，包括现有的神经网络和专门的特征/规则工程方法。我们还努力可视化基于树的卷积过程，来解释我们的模型工作原理。

基于树形卷积的辨别神经句子建模