A Hybrid Approach of Opinion Mining and Comparative Linguistic Analysis of Restaurant Reviews

Proceedings of Recent Advances in Natural Language Processing, pages 1281–1288 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_144 1281 A Hybrid Approach of Opinion Mining and Comparative Linguistic Analysis of Restaurant Reviews Salim Sazzed Old Dominion University Norfolk, VA, USA ssazz001@odu.edu Abstract The existing research on sentiment analysis mainly utilized data curated in limited geo- graphical regions and demography (e.g., USA, UK, China) due to commercial interest and availability of review data. Since the user’s at- titudes and preferences can be affected by nu- merous sociocultural factors and demographic characteristics, it is necessary to have anno- tated review datasets belong to various demog- raphy. In this work, we first construct a re- view dataset BanglaRestaurant that contains over 2300 customer reviews towards a number of Bangladeshi restaurants. Then, we present a hybrid methodology that yields improvement over the best performing lexicon-based and machine learning (ML) based classifier with- out using any labeled data. Finally, we investi- gate how the demography (i.e., geography and nativeness in English) of users affect the lin- guistic characteristics of the reviews by con- trasting two datasets, BanglaRestaurant and Yelp. The comparative results demonstrate the efficacy of the proposed hybrid approach. The data analysis reveals that demography plays an influential role in the linguistic aspects of re- views. 1 Introduction Sentiment analysis or opinion mining refers to the process of identifying opinions or sentiments ex- pressed (e.g., positive, negative) in a text docu- ment (Liu, 2012) . The lexicon-based method and machine learning (ML) based method are the two dominant approaches for opinion mining; although, their combinations have been also explored by the researchers. To evaluate the polarity of a piece of text, the lexicon-based methods rely on the senti- ment lexicon comprised of opinion-conveying pos- itive or negative terms and a set of rules (Turney, 2002; Sazzed, 2020b). For the lexicon-based meth- ods, the laborious steps of data labeling are not required. The supervised machine learning (ML) based approaches derive the relationship between features of the text segments and the opinions ex- pressed in the writing in a supervised fashion (Pang et al., 2002; Sazzed and Jayarathna, 2019). There- fore, supervised ML classifiers require a significant amount of annotated data for training a predictive model for determining the polarity of a text docu- ment. Labeling a large amount of text data is not only a challenging but also a tedious and costly pro- cess (Sazzed and Jayarathna, 2021; Sazzed, 2020a). Researchers also investigated the combinations of both lexicon-based and ML-based approaches to form a hybrid method (Kolchyna et al., 2015; Men- don et al., 2021; Sazzed, 2021). Mining customer opinions towards restaurants has attained popularity in recent years due to its impact on business growth and sustainability. Re- searchers performed a number of studies using the Yelp restaurant review and several other datasets. However, existing research on opinion mining in review datasets mainly focused on data curated in some specific demography (e.g., developed coun- tries or countries with large consumer markets) due to commercial interest and abundance of user- …

Content truncated to 500 words