Topic-Aware Neural Keyphrase Generation for Social Media Language

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2516–2526 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 2516 Topic-Aware Neural Keyphrase Generation for Social Media Language Yue Wang1∗ Jing Li2† Hou Pong Chan1 Irwin King1 Michael R. Lyu1 Shuming Shi2 1Department of Computer Science and Engineering The Chinese University of Hong Kong, HKSAR, China 2Tencent AI Lab, Shenzhen, China 1{yuewang, hpchan, king,lyu}@cse.cuhk.edu.hk 2{ameliajli,shumingshi}@tencent.com Abstract A huge volume of user-generated content is daily produced on social media. To facil- itate automatic language understanding, we study keyphrase prediction, distilling salient information from massive posts. While most existing methods extract words from source posts to form keyphrases, we propose a sequence-to-sequence (seq2seq) based neural keyphrase generation framework, enabling ab- sent keyphrases to be created. Moreover, our model, being topic-aware, allows joint mod- eling of corpus-level latent topic representa- tions, which helps alleviate the data sparsity that widely exhibited in social media language. Experiments on three datasets collected from English and Chinese social media platforms show that our model significantly outperforms both extraction and generation models that do not exploit latent topics.1 Further discussions show that our model learns meaningful topics, which interprets its superiority in social media keyphrase generation. 1 Introduction As social media continues its worldwide expan- sion, the last decade has witnessed the revolution of interpersonal communication. While empow- ering individuals with richer and fresher informa- tion, the flourish of social media also results in millions of posts generated on a daily basis. Fac- ing a sheer quantity of texts, language understand- ing has become a daunting task for human beings. Under this circumstance, there exists a pressing need for developing automatic systems capable of absorbing massive social media texts and figuring out what is important. *This work was partially done when Yue Wang was an intern at Tencent AI Lab. †Jing Li is the corresponding author. 1Our data and code are publicly released in https:// github.com/yuewang-cuhk/TAKG Source post with keyphrase “super bowl”: [S]: Somewhere, a wife that is not paying attention to the game, says ”I want the team in yellow pants to win.” Relevant tweets: [T1]: I been a steelers fan way before black & yel- low and this super bowl! [T2]: I will bet you the team with yellow pants wins. [T3]: Wiz Khalifa song ’black and yellow” to spur the pittsburgh steelers and Lil Wayne is to sing ”green and yellow’ for the packers. Table 1: Sample tweets tagged with “super bowl” as their keyphrases. Blue and italic words can indicate the topic of super bowl. In this work, we study the prediction of keyphrases, generally formed with words or phrases reflecting main topics conveyed in input texts (Zhang et al., 2018). Particularly, we fo- cus on producing keyphrases for social media lan- guage, proven to be beneficial to a broad range of applications, such as instant detection of trending events (Weng and Lee, 2011), summarizing pub- lic opinions (Meng et al., 2012), analyzing social behavior (Ruths …

Content truncated to 500 words