Microblog Hashtag Generation via Encoding Conversation Contexts

Proceedings of NAACL-HLT 2019, pages 1624–1633 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 1624 Microblog Hashtag Generation via Encoding Conversation Contexts Yue Wang 1∗ Jing Li2† 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, king, lyu}@cse.cuhk.edu.hk 2{ameliajli,shumingshi}@tencent.com Abstract Automatic hashtag annotation plays an im- portant role in content understanding for mi- croblog posts. To date, progress made in this field has been restricted to phrase selection from limited candidates, or word-level hash- tag discovery using topic models. Different from previous work considering hashtags to be inseparable, our work is the first effort to an- notate hashtags with a novel sequence gener- ation framework via viewing the hashtag as a short sequence of words. Moreover, to address the data sparsity issue in processing short mi- croblog posts, we propose to jointly model the target posts and the conversation contexts initi- ated by them with bidirectional attention. Ex- tensive experimental results on two large-scale datasets, newly collected from English Twit- ter and Chinese Weibo, show that our model significantly outperforms state-of-the-art mod- els based on classification.1 Further studies demonstrate our ability to effectively generate rare and even unseen hashtags, which is how- ever not possible for most existing methods. 1 Introduction Microblogs have become an essential outlet for in- dividuals to voice opinions and exchange infor- mation. Millions of user-generated messages are produced every day, far outpacing the human be- ing’s reading and understanding capacity. As a result, the current decade has witnessed the in- creasing demand for effectively discovering gist information from large microblog texts. To iden- tify the key content of a microblog post, hashtags, user-generated labels prefixed with a “#” (such as “#NAACL” and “#DeepLearning”), have been *This work was mainly done when Yue Wang was an in- tern at Tencent AI Lab. †Jing Li is the corresponding author. 1To obtain our datasets, please contact Yue Wang and Jing Li. Target post for hashtag generation This Azarenka woman needs a talking to from the umpire her weird noises are totes inappropes pro- fessionally. #AusOpen Replying messages forming a conversation [T1] How annoying is she. I just worked out what she sounds like one of those turbo charged cars when they change gear or speed. [T2] On the topic of noises, I was at the Nadal- Tomic game last night and I loved how quiet Tomic was compared to Nadal. [T3] He seems to have a shitload of talent and the postmatch press conf. He showed a lot of maturity and he seems nice. [T4] Tomic has a fantastic tennis brain... Table 1: A post and its conversation snippet about “Australian Open” on Twitter. “#AusOpen” is the human-annotated hashtag for the target post. Words in- dicative of the hashtag are in blue and italic type. widely used to reflect keyphrases (Zhang et al., 2016, 2018) or topics (Yan et al., 2013; Hong et al., 2012; …

Content truncated to 500 words