An Integrated Approach for Keyphrase Generation via Exploring the Power of Retrieval and Extraction

Proceedings of NAACL-HLT 2019, pages 2846–2856 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 2846 An Integrated Approach for Keyphrase Generation via Exploring the Power of Retrieval and Extraction Wang Chen1, Hou Pong Chan1, Piji Li2, Lidong Bing3, Irwin King1 1The Chinese University of Hong Kong, Shatin, N.T., Hong Kong 2Tencent AI Lab 3R&D Center Singapore, Machine Intelligence Technology, Alibaba DAMO Academy 1{wchen, hpchan, king}@cse.cuhk.edu.hk 2pijili@tencent.com, 3l.bing@alibaba-inc.com Abstract In this paper, we present a novel integrated ap- proach for keyphrase generation (KG). Unlike previous works which are purely extractive or generative, we first propose a new multi- task learning framework that jointly learns an extractive model and a generative model. Besides extracting keyphrases, the output of the extractive model is also employed to rec- tify the copy probability distribution of the generative model, such that the generative model can better identify important contents from the given document. Moreover, we re- trieve similar documents with the given doc- ument from training data and use their asso- ciated keyphrases as external knowledge for the generative model to produce more accu- rate keyphrases. For further exploiting the power of extraction and retrieval, we propose a neural-based merging module to combine and re-rank the predicted keyphrases from the enhanced generative model, the extractive model, and the retrieved keyphrases. Experi- ments on the five KG benchmarks demonstrate that our integrated approach outperforms the state-of-the-art methods. 1 Introduction Keyphrases are short text pieces that can quickly express the key ideas of a given document. The keyphrase generation task aims at automatically generating a set of keyphrases given a docu- ment. As shown in the upper part of Figure 1, the input is a document and the output is a set of keyphrases. Due to the concise and precise expression, keyphrases are beneficial to exten- sive downstream applications such as text sum- marization (Zhang et al., 2004; Wang and Cardie, 2013), sentiment analysis (Wilson et al., 2005; Berend, 2011), and document clustering (Hulth and Megyesi, 2006; Hammouda et al., 2005). Existing methods on keyphrase generation Document: Futility-Based Offspring Sizing. Parameter control in evolutionary algorithms (EAs) has been shown to be beneficial; however, the control of offspring size has so far received very little attention. This paper introduces Futility-Based Offspring Sizing (FuBOS), a method for controlling offspring size on a per generation basis without even requiring the user to set an initial offspring size value. . . Keyphrases: {evolutionary algorithm; parameterless evolutionary algorithm; parameter control; offspring sizing; optimization} Retrieved Document: An Exploration into Dynamic Population Sizing. Traditional evolutionary algorithms are powerful problem solvers that have several fixed parameters which require prior specification. . . While many methods of parameter control have been published that focus on removing the population size parameter, µ, all hampered by a variety of problems. This paper investigates the benefits of making µ a dynamic parameter and introduces two novel methods for population control. . . Retrieved Keyphrases: {evolutionary algorithm; parameterless evolutionary algorithm; parameter control; population sizing; optimization} Figure …

Content truncated to 500 words