A Large-Scale Japanese Dataset for Aspect-based Sentiment Analysis

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 7014–7021 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 7014 A Large-Scale Japanese Dataset for Aspect-based Sentiment Analysis Yuki Nakayama, Koji Murakami, Gautam Kumar, Sudha Bhingardive, Ikuko Hardaway Rakuten Institute of Technology, Rakuten Group Inc. Tokyo, Japan {yuki.b.nakayama, koji.murakami, gautam.kumar, sudha.bhingardive, ikuko.hardaway}@rakuten.com Abstract There has been significant progress in the field of sentiment analysis. However, aspect-based sentiment analysis (ABSA) has not been explored in the Japanese language even though it has a huge scope in many natural language processing applications such as 1) tracking sentiment towards products, movies, politicians etc; 2) improving customer relation models. The main reason behind this is that there is no standard Japanese dataset available for ABSA task. In this paper, we present the first standard Japanese dataset for the hotel reviews domain. The proposed dataset contains 53,192 review sentences with seven aspect categories and two polarity labels. We perform experiments on this dataset using popular ABSA approaches and report error analysis. Our experiments show that contextual models such as BERT works very well for the ABSA task in the Japanese language and also show the need to focus on other NLP tasks for better performance through our error analysis. Keywords: Aspect-based Sentiment Analysis, Dataset Construction, Less-Resourced 1. Introduction Sentiment analysis is a very well-known and essential field in NLP. It is used to identify the sentiment of a text or text span and classify it into pre-defined categories like positive, negative, neutral, etc. Aspect-based sen- timent analysis is a fine-grained task in sentiment anal- ysis. In this task, it is necessary to recognize words representing “aspect” in addition to identifying the po- larity. For example, hotel reviews not only express the overall sentiment of a meal, but also sentiments re- garding its service, atmosphere, location, amenity, and so on. Let us consider a hotel review sentence in the Japanese language with the following example: “朝食は美味しかったですが、スタッフのサービ スはいまいちでした。” (Breakfast was delicious, but the service of the staff was not so good.) In this sentence, there are two aspects: 朝食 (break- fast) and サービス (service). It has positive polar- ity with respect to breakfast but negative polarity re- garding service. In many businesses, it is indispens- able to analyze such fine-grained polarities concern- ing different aspects to improve the quality of services. There are many versions of aspect-based sentiment analysis (ABSA) tasks which are aspect-category sen- timent analysis (ACSA), aspect-term sentiment anal- ysis (ATSA), and targeted-aspect sentiment analysis (TABSA). In sentiment analysis literature, there has been various approaches proposed so far on various languages including Japanese. For research on ABSA, several datasets have been constructed, including SemEval-2014 Restaurant Re- view dataset, Laptop Review dataset (Pontiki et al., 2014), Twitter (Dong et al., 2014), SentiHood (Saeidi et al., 2016), SemEval-2015 (Pontiki et al., 2015) and another large-scale Multi-Aspect Multi-Sentiment (MAMS) dataset (Jiang et al., 2019). These datasets are based on GGM (Consumer Generated Media) and have become the benchmark datasets for …

Content truncated to 500 words