A New Direction in Stance Detection: Target-Stance Extraction in the Wild

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 10071–10085 July 9-14, 2023 ©2023 Association for Computational Linguistics A New Direction in Stance Detection: Target-Stance Extraction in the Wild Yingjie Li∗ Krishna Garg∗ Cornelia Caragea University of Illinois at Chicago {yli300,kgarg8,cornelia}@uic.edu Abstract Stance detection aims to detect the stance to- ward a corresponding target. Existing works have achieved promising progress on stance de- tection tasks in which the goal is to predict the stance given both a target and a text. However, they all work under the assumption that the tar- get is known in advance, which is often not the case in the wild. Given a text from social media platforms, the target information is often unknown due to implicit mentions in the source text and it is infeasible to have manual target annotations at a large scale. Therefore, in this paper, we propose a new task Target-Stance Ex- traction (TSE) that aims to extract the (target, stance) pair from the text. We benchmark the task by proposing a two-stage framework that first identifies the relevant target in the text and then detects the stance given the predicted tar- get and text. Specifically, we first propose two different settings: Target Classification and Tar- get Generation, to identify the potential target from a given text. Then we propose a multi- task approach that takes target prediction as the auxiliary task to detect the stance toward the predicted target. We evaluate the proposed framework on both in-target stance detection in which the test target is always seen in the training stage and zero-shot stance detection that needs to detect the stance for the unseen target during the inference stage. The new TSE task can facilitate future research in the field of stance detection. We publicly release our code.1 1 Introduction Stance detection aims to automatically identify peo- ple’s attitude/viewpoint (e.g., Favor or Against) ex- pressed in texts toward a target that is generally a controversial topic or political figure (ALDayel and Magdy, 2021; Küçük and Can, 2020; Hardalov et al., 2021). For example, the tweet in Figure 1 ∗Both authors contributed equally to this research. 1https://github.com/chuchun8/TSE Figure 1: The comparison between the proposed Target- Stance Extraction (TSE) task and original stance detec- tion task. expresses a stance of “Against” toward the target “Atheism.” Social media platforms like Twitter, Facebook and other debate forums have become an integral way of opinion dissemination these days (Khan et al., 2021). The peculiar characteristics of these platforms are that the information is usually scat- tered across texts and the opinionated text could be expressed toward target entities in an implicit way. Existing methods have achieved promising performance on in-target stance detection in which same targets are seen in both train and test sets (Mohammad et al., 2016a; Sobhani et al., 2017; Li and Caragea, 2019, 2021a) and cross-target stance detection that aims to transfer the knowledge from a source target to a destination target (Augenstein et al., …

Content truncated to 500 words