Multi-Domain Targeted Sentiment Analysis

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2751 - 2762 July 10-15, 2022 ©2022 Association for Computational Linguistics Multi-Domain Targeted Sentiment Analysis Orith Toledo-Ronen, Matan Orbach, Yoav Katz, Noam Slonim IBM Research {oritht, matano, katz, noams}@il.ibm.com Abstract Targeted Sentiment Analysis (TSA) is a central task for generating insights from consumer re- views. Such content is extremely diverse, with sites like Amazon or Yelp containing reviews on products and businesses from many differ- ent domains. A real-world TSA system should gracefully handle that diversity. This can be achieved by a multi-domain model – one that is robust to the domain of the analyzed texts, and performs well on various domains. To address this scenario, we present a multi-domain TSA system based on augmenting a given training set with diverse weak labels from assorted do- mains. These are obtained through self-training on the YELP reviews corpus. Extensive exper- iments with our approach on three evaluation datasets across different domains demonstrate the effectiveness of our solution. We further analyze how restrictions imposed on the avail- able labeled data affect the performance, and compare the proposed method to the costly al- ternative of manually gathering diverse TSA labeled data. Our results and analysis show that our approach is a promising step towards a practical domain-robust TSA system. 1 Introduction Customer reviews of products and businesses pro- vide insights for both consumers and companies. They help companies understand customer satisfac- tion or guide marketing campaigns, and aid con- sumers in their decision-making. Sentiment anal- ysis plays a central role in the analysis of such material, by aiming to understand the sentiment expressed in a review document or in a single re- view sentence (Liu, 2012). Beyond these high-level trends, identifying the sentiment towards a specific product feature or an entity is important. Such a fine-grained analysis includes the key task of Tar- geted Sentiment Analysis (TSA), aimed at detect- ing sentiment-bearing terms in texts and classifying the sentiment towards them. For example, in the sentence "The room was noisy, but the food was tasty," the targets are room and food with negative and positive sentiments, respectively. Our focus in this work is on TSA of user reviews in English. A real-world TSA system has to successfully process diverse data. From toothbrushes to phones, airline companies to local retailers, the online con- tent today covers a broad range of reviews in many domains. Ideally, a system for such a multi-domain scenario should be able to cope with inputs from any domain, those that were seen during training, and, perhaps more importantly, those that were not. To the best of our knowledge, this work is the first to pursue TSA in a multi-domain setup, in- tending to support input from multiple unknown domains. Many previous works have used the in- domain setup of training and testing on data from the same domain (e.g. Li et al. (2019b)). Newer works focus on the cross-domain setup, …

Content truncated to 500 words