Collection and Analysis of Travel Agency Task Dialogues with Age-Diverse Speakers

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 574–580 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 574 Aspect-Based Emotion Analysis and Multimodal Coreference: A Case Study of Customer Comments on Adidas Instagram Posts Luna De Bruyne1,∗, Akbar Karimi2,∗, Orph´ee De Clercq1, Andrea Prati2, V´eronique Hoste1 1LT3, Language and Translation Technology Team, Ghent University Groot-Brittannielaan 45, 9000 Ghent, Belgium 2IMP Lab, University of Parma Parco Area delle Scienze, 181/A, 43124 Parma, Italy luna.debruyne@ugent.be, akbar.karimi@unipr.it, orphee.declercq@ugent.be, andrea.prati@unipr.it, veronique.hoste@ugent.be Abstract While aspect-based sentiment analysis of user-generated content has received a lot of attention in the past years, emotion detection at the aspect level has been relatively unexplored. Moreover, given the rise of more visual content on social media platforms, we want to meet the ever-growing share of multimodal content. In this paper, we present a multimodal dataset for Aspect-Based Emotion Analysis (ABEA). Additionally, we take the first steps in investigating the utility of multimodal coreference resolution in an ABEA framework. The presented dataset consists of 4,900 comments on 175 images and is annotated with aspect and emotion categories and the emotional dimensions of valence and arousal. Our preliminary experiments suggest that ABEA does not benefit from multimodal coreference resolution, and that aspect and emotion classification only requires textual information. However, when more specific information about the aspects is desired, image recognition could be essential. Keywords: ABSA, ABEA, Sentiment Analysis, Emotion Detection, Multimodal Coreference 1. Introduction Nowadays, all large companies and brands own a des- ignated page or channel on social media platforms which allows them to directly communicate with var- ious stakeholders, most notably the general public. This rise of social media communication has also propagated research on automatically analysing user- generated content which is often full of opinions and emotions. As a consequence, research on sentiment and emotion analysis has thrived. Since companies not only want to know which gen- eral opinions their stakeholders hold on their com- pany or brand, but also want to learn which products or which features of their products are greatly appre- ciated or disliked by the online community, the re- search domain of sentiment analysis has started to fo- cus more and more on fine-grained sentiment analysis (positive/negative/neutral) at the feature or aspect level (Liu, 2012). This is referred to as Aspect-Based Sen- timent Analysis (ABSA), which focuses on the detec- tion of all sentiment expressions within a given doc- ument and the concepts and aspects (or features) to which they refer. Following the task description in the SemEval 2015 shared task on this topic (Pontiki et al., 2015), aspect-based sentiment analysis can be de- composed into three subtasks: aspect term extraction, aspect category classification and aspect term polarity classification. In its original sense, sentiment analysis involves clas- sifying instances as one of the three classes positive, negative, and neutral, but more recently, many research ∗Equal contribution objectives shifted to extracting more fine-grained emo- tional information like the emotional categories anger, sadness, and joy (Mohammad, …

Content truncated to 500 words