Taxonomy Construction of Unseen Domains via Graph-based Cross-Domain Knowledge Transfer

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 2198–2208 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 2198 Taxonomy Construction of Unseen Domains via Graph-based Cross-Domain Knowledge Transfer Chao Shang1, Sarthak Dash2, Md Faisal Mahbub Chowdhury2, Nandana Mihindukulasooriya2, Alfio Gliozzo2 1University of Connecticut, Storrs, CT, USA 2IBM Research AI, Yorktown Heights, NY, USA chao.shang@uconn.edu, sdash@us.ibm.com mchowdh@us.ibm.com, nandana.m@ibm.com, gliozzo@us.ibm.com Abstract Extracting lexico-semantic relations as graph- structured taxonomies, also known as taxon- omy construction, has been beneficial in a vari- ety of NLP applications. Recently Graph Neu- ral Network (GNN) has shown to be power- ful in successfully tackling many tasks. How- ever, there has been no attempt to exploit GNN to create taxonomies. In this paper, we pro- pose Graph2Taxo, a GNN-based cross-domain transfer framework for the taxonomy construc- tion task. Our main contribution is to learn the latent features of taxonomy construction from existing domains to guide the structure learning of an unseen domain. We also pro- pose a novel method of directed acyclic graph (DAG) generation for taxonomy construction. Specifically, our proposed Graph2Taxo uses a noisy graph constructed from automatically extracted noisy hyponym-hypernym candidate pairs, and a set of taxonomies for some known domains for training. The learned model is then used to generate taxonomy for a new un- known domain given a set of terms for that domain. Experiments on benchmark datasets from science and environment domains show that our approach attains significant improve- ments correspondingly over the state of the art. 1 Introduction Taxonomy has been exploited in many Natural Lan- guage Processing (NLP) applications, such as ques- tion answering (Harabagiu et al., 2003), query un- derstanding (Hua et al., 2017), recommendation systems (Friedrich and Zanker, 2011), etc. Auto- matic taxonomy construction is highly challenging as it involves the ability to recognize – (i) a set of types (i.e. hypernyms) from a text corpus, (ii) instances (i.e. hyponyms) of each type, and (iii) is-a (i.e. hypernymy) hierarchy between types. Existing taxonomies (e.g., WordNet (Miller et al., 1990)) are far from being complete. Tax- onomies specific to many domains are either en- tirely absent or missing. In this paper, we focus on construction of taxonomies for such unseen do- mains1. Since taxonomies are expressed as directed acyclic graphs (DAGs) (Suchanek et al., 2008), tax- onomy construction can be formulated as a DAG generation problem. There has been considerable research on Graph Neural Networks (GNN) (Sperduti and Starita, 1997; Gori et al., 2005) over the years; particu- larly inspired by the convolutional GNN (Bruna et al., 2014) where graph convolution operations were defined in the Fourier domain. In a similar spirit to convolutional neural networks (CNNs), GNN methods aggregate neighboring information based on the connectivity of the graph to create node embeddings. GNN has been applied suc- cessfully in many tasks such as matrix comple- tion (van den Berg et al., 2017), manifold analy- sis (Monti et al., 2017), predictions of community (Bruna et al., 2014), knowledge graph completion (Shang et al., …

Content truncated to 500 words