This chapter applies tree-based convolution to the dependency parse trees of natural language sentences, resulting in a new variant d-TBCNN. Since dependency trees are different from abstract syntax trees in Chap. 4 and constituency trees in Chap. 5, we need to design new model gadgets for d-TBCNN. The model is evaluated on two sentence classification tasks (sentiment analysis and question classification) and a sentence matching task. In the sentence classification tasks, d-TBCNN outperforms previous state-of-the-art results, whereas in the sentence matching task, d-TBCNN achieves comparable performance to the previous state-of-the-art model, which has a higher matching complexity.
CITATION STYLE
Mou, L., & Jin, Z. (2018). TBCNN for dependency trees in natural language processing. In SpringerBriefs in Computer Science (pp. 73–89). Springer. https://doi.org/10.1007/978-981-13-1870-2_6
Mendeley helps you to discover research relevant for your work.