Hashtags, Emotions, and Comments: A Large-Scale Dataset to Understand Fine-Grained Social Emotions to Online Topics

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 1376–1382, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 1376 Hashtags, Emotions, and Comments: A Large-Scale Dataset to Understand Fine-Grained Social Emotions to Online Topics Keyang Ding Jing Li∗ Yuji Zhang Department of Computing, The Hong Kong Polytechnic University, HKSAR, China {keyang.ding,yu-ji.zhang}@connect.polyu.hk jing-amelia.li@polyu.edu.hk Abstract This paper studies social emotions to online discussion topics. While most prior work fo- cus on emotions from writers, we investigate readers’ responses and explore the public feel- ings to an online topic. A large-scale dataset is collected from Chinese microblog Sina Weibo with over 13 thousand trending topics, emo- tion votes in 24 fine-grained types from mas- sive participants, and user comments to al- low context understanding.1 In experiments, we examine baseline performance to predict a topic’s possible social emotions in a multi- label classification setting. The results show that a seq2seq model with user comment mod- eling performs the best, even surpassing hu- man prediction. More analyses shed light on the effects of emotion types, topic description lengths, contexts from user comments, and the limited capacity of the existing models. 1 Introduction Social media have become a popular outlet for peo- ple to voice opinions, share viewpoints, and ex- change ideas. It provides us with rich resource to research public opinions of the trending topics and hear people’s voice over important social events, such as the global COVID-19 crisis. However, our ever-changing physical world leads to the rapid evolution of discussion topics in online world; it is far beyond humans’ capability to catch them in real time. It consequently presents a pressing need for automatic sentiment (Wang et al., 2011; Yang and Eisenstein, 2017) and emotion (Abdul-Mageed and Ungar, 2017; Arag´on et al., 2019) analysis. Nevertheless, most of the related work focus on the feelings from writers (Tang et al., 2014; Huang and Carley, 2019; Singh et al., 2019) and ∗Jing Li is the corresponding author. 1The dataset can be found at: https://github.com/ polyusmart/HEC-Dataset [H]:#张艺兴整蛊GAI# [T]: Lay played tricks on GAI. [E]: : lol; : facepalm; : doge (tease). Figure 1: A Weibo hashtag and its resulting social emo- tions. H is the original hashtag in Chinese and T is its English translation. E shows the top three emojis voted by online users and their meanings (seperated by “:”). the existing studies concerning reader emotions mostly tackle well-written texts, such as news re- ports (Li et al., 2016; Zhou et al., 2018; Yang et al., 2019). Limited work has been done to character- ize collective feelings from the public (henceforth social emotions) to an online topic described with fragmented and colloquial social media language. Where some previous efforts gather viewpoints from limited readers through user replies (Alhothali and Hoey, 2015; Li et al., 2019) or manual anno- tations (Buechel and Hahn, 2017; Bostan et al., 2019), we focus on social emotions reflecting ag- gregated feelings from large amount of people. In light of these concerns, we present a novel task to …

Content truncated to 500 words