Structure-Augmented Keyphrase Generation

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 2657–2667 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 2657 Structure-Augmented Keyphrase Generation Jihyuk Kim Yonsei University jihyukkim@yonsei.ac.kr Myeongho Jeong Yonsei University wag9611@yonsei.ac.kr Seungtaek Choi Yonsei University hist0613@yonsei.ac.kr Seung-won Hwang∗ Seoul National University seungwonh@snu.ac.kr Abstract This paper studies the keyphrase generation (KG) task for scenarios where structure plays an important role. For example, a scien- tific publication consists of a short title and a long body, where the title can be used for de-emphasizing unimportant details in the body. Similarly, for short social media posts (e.g., tweets), scarce context can be augmented from titles, though often missing. Our con- tribution is generating/augmenting structure then encoding these information, using exist- ing keyphrases of other documents, comple- menting missing/incomplete titles. Specifi- cally, we first extend the given document with related but absent keyphrases from existing keyphrases, to augment missing contexts (gen- erating structure), and then, build a graph of keyphrases and the given document, to ob- tain structure-aware representation of the aug- mented text (encoding structure). Our empir- ical results validate that our proposed struc- ture augmentation and structure-aware encod- ing can improve KG for both scenarios, out- performing the state-of-the-art1. 1 Introduction Keyphrases not only help human readers to gain im- mediate insights about a given document, but also make documents queryable, e.g., by hashtagging social posts with predicted keyphrases. Among keyphrase tasks, we focus on tasks that allow words that are not present in the given document, which often be called absent words, to be keyphrases, as the input document can be often context-scarce: Many potentially relevant keyphrases are missing in the given document, since the keyphrase can be expressed in terms different from those present in the document (i.e., vocabulary mismatch). Our goal is to generate keyphrases that are likely to be words absent from the document, especially ∗Corresponding author. 1Our code is available at https://github.com/ jihyukkim-nlp/StrAugKG. Field Text Title Methods to tell if a question can be answered from a paragraph Question ... Is there any existing research in telling whether or not a question is answerable ? ... ... I considered textual entailment, but it doesn’t seem to be exactly what I’m looking for ... Existing ..., “natural language processing”, Keyphrases “formal languages”, ... Gold “question answering”, Keyphrases “natural language processing” Table 1: An example of a social Q&A post consisting of a title and the main body question: We present gold keyphrases labeled by the user, and existing keyphrases labeled for other related posts. Bold-face words can indicate the topic of the post. in the following scenarios: (a) scientific publica- tions with diverse vocabularies, or (b) short social posts, where we found that, from public and real- life datasets we used in our evaluation, 37% of keyphrases in scientific publications, 67% of hash- tags, on average, in social media posts are absent in the given document. Thus, we formulate our prob- lem as keyphrase generation (KG) (Meng et al., 2017) of predicting keyphrases, …

Content truncated to 500 words