A Multi-Task Learning Framework for Multi-Target Stance Detection

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 2320–2326 August 1–6, 2021. ©2021 Association for Computational Linguistics 2320 A Multi-Task Learning Framework for Multi-Target Stance Detection Yingjie Li and Cornelia Caragea Computer Science Department University of Illinois at Chicago yli300@uic.edu, cornelia@uic.edu Abstract Multi-target stance detection aims to identify the stance taken toward a pair of different tar- gets from the same text, and typically, there are multiple target pairs per dataset. Existing works generally train one model for each target pair. However, they fail to learn target-specific representations and are prone to overfitting. In this paper, we propose a new training strategy under the multi-task learning setting by train- ing one model on all target pairs, which helps the model learn more universal representations and alleviate overfitting. Moreover, in order to extract more accurate target-specific represen- tations, we propose a multi-task learning net- work which can jointly train our model with a stance (dis)agreement detection task that is de- signed to identify agreement and disagreement between stances in paired texts. Experimental results demonstrate that our proposed model outperforms the best-performing baseline by 12.39% in macro-averaged F1-score. Our re- sources are publicly available on GitHub.1 1 Introduction Nowadays, people often take to social media to express their stances toward specific targets (e.g., various political figures). These stances in an aggre- gate can provide valuable information for obtaining insight into important events such as presidential elections. The common stance detection task is to determine from a piece of text whether the author of the text is in favor of, neutral or against to a specific target, which can be categorized as single- target stance detection (STSD) (Mohammad et al., 2016; K¨uc¸¨uk and Can, 2020; ALDayel and Magdy, 2021). More recently, since people often comment on multiple target entities in the same text, a more challenging task, i.e., multi-target stance detection (MTSD), was designed to test whether a model can 1https://github.com/chuchun8/MTSD Tweet: #Trump2016 can beat #HillaryClinton as he is easily beating #JebBush ;) People R sick and tired of career politicians. Target 1: Target 2: Stance 1: Stance 2: Donald Trump Hillary Clinton FAVOR AGAINST Figure 1: An example of multi-target stance detection. Target 1 Target Pair 1 Model1 Target Pair 2 Target Pair 3 Target 2 Target 3 Target 4 Target 5 Target 6 Model2 Model3 (a) Model Target Pair 1 Target Pair 2 Target Pair 3 Target 1 Target 2 Target 3 Target 4 Target 5 Target 6 (b) Figure 2: The left figure represents the previous “Ad- hoc” training setting in which a model is trained only on a target pair. The right figure represents the pro- posed “Merged” training setting in which the model is trained on all target pairs. accurately predict the stance toward multiple tar- gets in the same text (Sobhani et al., 2017). For example, for the tweet in Figure 1, the author aims at expressing stance toward two targets, Donald Trump and Hillary Clinton, implied by the pres- ence of the words “beat” …

Content truncated to 500 words