SGG: Learning to Select, Guide, and Generate for Keyphrase Generation

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5717–5726 June 6–11, 2021. ©2021 Association for Computational Linguistics 5717 SGG: Learning to Select, Guide, and Generate for Keyphrase Generation Jing Zhao, Junwei Bao, Yifan Wang, Youzheng Wu, Xiaodong He, Bowen Zhou JD AI Research {zhaojing857,baojunwei,wangyifan15,wuyouzheng1}@jd.com {xiaodong.he,bowen.zhou}@jd.com Abstract Keyphrases, that concisely summarize the high-level topics discussed in a document, can be categorized into present keyphrase which explicitly appears in the source text, and ab- sent keyphrase which does not match any con- tiguous subsequence but is highly semantically related to the source. Most existing keyphrase generation approaches synchronously gener- ate present and absent keyphrases without ex- plicitly distinguishing these two categories. In this paper, a Select-Guide-Generate (SGG) ap- proach is proposed to deal with present and absent keyphrase generation separately with different mechanisms. Specifically, SGG is a hierarchical neural network which consists of a pointing-based selector at low layer con- centrated on present keyphrase generation, a selection-guided generator at high layer ded- icated to absent keyphrase generation, and a guider in the middle to transfer informa- tion from selector to generator. Experimen- tal results on four keyphrase generation bench- marks demonstrate the effectiveness of our model, which significantly outperforms the strong baselines for both present and absent keyphrases generation. Furthermore, we ex- tend SGG to a title generation task which indi- cates its extensibility in natural language gen- eration tasks. 1 Introduction Automatic keyphrase prediction recommends a set of representative phrases that are related to the main topics discussed in a document (Liu et al., 2009). Since keyphrases can provide a high-level topic description of a document, they are beneficial for a wide range of natural language processing (NLP) tasks, such as information extraction (Wan and Xiao, 2008), text summarization (Wang and Cardie, 2013) and question generation (Subrama- nian et al., 2018). <title> enhancing product recommender systems on sparse binary data </title> <abstract>… rank the recommendations online for the customer . the second phase differs from the traditional approach and an empirical comparison between the methods used in e-VZpro and other collaborative filtering methods including dependency networks …</abstract> customer relationship management collaborative filtering <Present keyphrase> </Present keyphrase> <Absent keyphrase> </Absent keyphrase> Selection-Guided Generation Select Figure 1: An example of keyphrase prediction by SGG. Existing methods for keyphrase prediction can be categorized into extraction and generation ap- proaches. Specifically, keyphrase extraction meth- ods identify important consecutive words from a given document as keyphrases, which means that the extracted keyphrases (denoted as present keyphrases) must exactly come from the given doc- ument. However, some keyphrases (denoted as absent keyphrases) of a given document do not match any contiguous subsequence but are highly semantically related to the source text. The extrac- tion methods fail to predict these absent keyphrases. Therefore, generation methods have been proposed to produce a keyphrase verbatim from a prede- fined vocabulary, no matter whether the gener- ated keyphrase appears in the source text. Com- pared with conventional extraction methods, …

Content truncated to 500 words