Neural Keyphrase Generation via Reinforcement Learning with Adaptive Rewards

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2163–2174 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 2163 Neural Keyphrase Generation via Reinforcement Learning with Adaptive Rewards Hou Pong Chan1, Wang Chen1, Lu Wang2, and Irwin King1 1The Chinese University of Hong Kong, Shatin, N.T., Hong Kong 2Northeastern Univesity, Boston, MA, USA 1{hpchan, wchen, king}@cse.cuhk.edu.hk 2luwang@ccs.neu.edu Abstract Generating keyphrases that summarize the main points of a document is a fundamen- tal task in natural language processing. Al- though existing generative models are capa- ble of predicting multiple keyphrases for an input document as well as determining the number of keyphrases to generate, they still suffer from the problem of generating too few keyphrases. To address this problem, we propose a reinforcement learning (RL) approach for keyphrase generation, with an adaptive reward function that encourages a model to generate both sufficient and accu- rate keyphrases. Furthermore, we introduce a new evaluation method that incorporates name variations of the ground-truth keyphrases us- ing the Wikipedia knowledge base. Thus, our evaluation method can more robustly evaluate the quality of predicted keyphrases. Exten- sive experiments on five real-world datasets of different scales demonstrate that our RL ap- proach consistently and significantly improves the performance of the state-of-the-art gener- ative models with both conventional and new evaluation methods. 1 Introduction The task of keyphrase generation aims at predict- ing a set of keyphrases that convey the core ideas of a document. Figure 1 shows a sample doc- ument and its keyphrase labels. The keyphrases in red color are present keyphrases that appear in the document, whereas the blue ones are absent keyphrases that do not appear in the input. By distilling the key information of a document into a set of succinct keyphrases, keyphrase genera- tion facilitates a wide variety of downstream ap- plications, including document clustering (Ham- mouda et al., 2005; Hulth and Megyesi, 2006), opinion mining (Berend, 2011), and summariza- tion (Zhang et al., 2004; Wang and Cardie, 2013). Document: DCE MRI data analysis for cancer area classification. The paper aims at improving the support of medical researchers in the context of in-vivo cancer imaging… The proposed approach is based on a three-step procedure: i) robust feature extraction from raw time-intensity curves, ii) voxel segmentation, and iii) voxel classification based on a learning-by-example approach… Finally, in the third step, a support vector machine (SVM) is trained to classify voxels according to the labels obtained by the clustering phase… Keyphrase labels: svm; dce mri; cluster analysis; classification catSeqD predictions: cancer area classification; support vector machine catSeqD-𝟏���𝑹𝑭𝟏 predictions: dce mri; cancer area classification; support vector machine; image segmentation; morphological analysis Enriched keyphrase labels: {svm, support vector machine}; dce mri; cluster analysis; classification Figure 1: Sample document with keyphrase labels and predicted keyphrases. We use red (blue) color to highlight present (absent) keyphrases. The under- lined phrases are name variations of a keyphrase label. “catSeqD” is a keyphrase generation model from Yuan et al. (2018). “catSeqD-2RF1” …

Content truncated to 500 words