Exclusive Hierarchical Decoding for Deep Keyphrase Generation

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 1095–1105 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 1095 Exclusive Hierarchical Decoding for Deep Keyphrase Generation Wang Chen1, Hou Pong Chan1, Piji Li2, Irwin King1 1The Chinese University of Hong Kong, Shatin, N.T., Hong Kong 2Tencent AI Lab 1{wchen, hpchan, king}@cse.cuhk.edu.hk 2pijili@tencent.com Abstract Keyphrase generation (KG) aims to summa- rize the main ideas of a document into a set of keyphrases. A new setting is recently in- troduced into this problem, in which, given a document, the model needs to predict a set of keyphrases and simultaneously deter- mine the appropriate number of keyphrases to produce. Previous work in this setting em- ploys a sequential decoding process to gen- erate keyphrases. However, such a decoding method ignores the intrinsic hierarchical com- positionality existing in the keyphrase set of a document. Moreover, previous work tends to generate duplicated keyphrases, which wastes time and computing resources. To overcome these limitations, we propose an exclusive hi- erarchical decoding framework that includes a hierarchical decoding process and either a soft or a hard exclusion mechanism. The hierarchical decoding process is to explicitly model the hierarchical compositionality of a keyphrase set. Both the soft and the hard ex- clusion mechanisms keep track of previously- predicted keyphrases within a window size to enhance the diversity of the generated keyphrases. Extensive experiments on multi- ple KG benchmark datasets demonstrate the ef- fectiveness of our method to generate less du- plicated and more accurate keyphrases1. 1 Introduction Keyphrases are short phrases that indicate the core information of a document. As shown in Figure 1, the keyphrase generation (KG) problem focuses on automatically producing a keyphrase set (a set of keyphrases) for the given document. Because of the condensed expression, keyphrases can benefit various downstream applications including opinion mining (Berend, 2011; Wilson et al., 2005), doc- 1Our code is available at https://github.com/ Chen-Wang-CUHK/ExHiRD-DKG. Input Document: … A noninvasive diagnostic device was developed to assess the vascular origin and severity of penile dysfunction. It was designed and studied using both a mathematical model of penile hemodynamics and preliminary experiments on healthy young volunteers. … Simulations using a mathematical model show that the device is capable of differentiating between arterial insufficiency and venous leak and indicate the severity of each. … Keyphrases: {erectile dysfunction; arterial insufficiency; venous leak; veno-occlusive mechanism; mathematical model; hemodynamics} Figure 1: An example of an input document and its ex- pected keyphrase output for keyphrase generation prob- lem. Present keyphrases that appear in the document are underlined. ument clustering (Hulth and Megyesi, 2006), and text summarization (Wang and Cardie, 2013). Keyphrases of a document can be categorized into two groups: present keyphrase that appears in the document and absent keyphrase that does not appear in the document. Recent generative methods for KG apply the attentional encoder- decoder framework (Luong et al., 2015; Bahdanau et al., 2014) with copy mechanism (Gu et al., 2016; See et al., 2017) to predict both present …

Content truncated to 500 words