Evaluating Methods for Extraction of Aspect Terms in Opinion Texts in Portuguese - the Challenges of Implicit Aspects

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3819–3828 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3819 Evaluating Methods for Extraction of Aspect Terms in Opinion Texts in Portuguese – the Challenges of Implicit Aspects Mateus Tarcinalli Machado, Thiago Alexandre Salgueiro Pardo Interinstitutional Center for Computational Linguistics (NILC), Institute of Mathematical and Computer Sciences, University of S˜ao Paulo mateusmachado@usp.br, taspardo@icmc.usp.br Abstract One of the challenges of aspect-based sentiment analysis is the implicit mention of aspects. These are more difficult to identify and may require world knowledge to do so. In this work, we evaluate frequency-based, hybrid, and machine learning methods, including the use of the pre-trained BERT language model, in the task of extracting aspect terms in opinionated texts in Portuguese, emphasizing the analysis of implicit aspects. Besides the comparative evaluation of methods, the differential of this work lies in the analysis’s novelty using a typology of implicit aspects that shows the knowledge needed to identify each implicit aspect term, thus allowing a mapping of the strengths and weaknesses of each method. Keywords: Sentiment analysis, Aspect extraction, Implicit aspects 1. Introduction With the great expansion of social networks and e- commerce services and, consequently, the increase in the production of online reviews and comments, there was the need to develop new methods for processing such data. Sentiment analysis, or opinion mining, can be defined as an area of study at the intersection be- tween Computer Science and Linguistics that aims to automatically determine sentiments in a text (Taboada, 2016) in order to analyze people’s opinions, feelings, assessments, attitudes and emotions concerning prod- ucts, services, organizations, individuals, issues, topics, and their attributes (Liu, 2010). Sentiment analysis can be performed at three levels: document, sentence and aspect levels. Each of them includes different tasks and challenges, and the aspect- based sentiment analysis (ABSA) can be considered the most refined level. For example, consider the sen- tence “The hotel room was very small”. In this case, we have a negative opinion about the “room” aspect of the entity “hotel”. So we have a triple formed by the entity (hotel), aspect (room), and sentiment (negative). In ABSA, we can still have information about the au- thor and time when the opinion was issued, forming a quintuple (Liu, 2010). We can divide ABSA into several tasks, as those presented in the International Workshop on Semantic Evaluation (SemEval) (Pontiki et al., 2014): • Aspect term extraction: aims to identify terms present in a sentence that are related to an aspect. In the example, we have the word “room” related to the aspect “room”. • Aspect term polarity: aims to find the sentiment related to an aspect term, usually positive, nega- tive or neutral. • Aspect category detection: aims to identify the category of the aspect mentioned in a sentence. For example, we can find other aspect terms re- lated to the “room” category, such as “apartment” or “dormitory”. • Aspect category polarity: aims to detect …

Content truncated to 500 words