Article Preview
Top1. Introduction
The popularity of mobile commerce (m-commerce) has created many new opportunities for online retailers, especially for those small and medium-sized enterprises, to connect with their potential customers directly without going through unnecessary third-party intermediaries. With the rapid growth of m-commerce, an overwhelming amount of user reviews are posted daily (Liu et al., 2017) and available online for making purchase decisions. These reviews always contain some sentiment words. The strengths of the sentiment words can be further assessed to support marketing analysis and new product development (Hu and Liu, 2004). Online retailers can evaluate the level of customer satisfaction with their products and design suitable marketing activities by analyzing the reviews (Çalıa & Balaman, 2019). Various approaches integrated with text analysis to study the consumer sentiments have been developed for evaluating customer satisfaction (Xiao et. al., 2016; Siering et. al., 2018; Bi et. al., 2019), ranking product alternatives (Liu et al., 2017; Fan et. al., 2018; Ng & Law, 2020), and for summarizing a number of text reviews in social networking sites with the inclusion of automatic text classification models (Mamkis and Malamos, 2011; Ali et al., 2017). Sentiment Analysis, also known as opinion mining, is the use of text analysis to identify or study the subjective information of the user-generated content. Sentiment analysis can be carried out by searching sentiment words based on different lexical resources to determine the contextual polarity of a sentence (Hu & Liu, 2004; Guerini, et. al., 2013; Cambria, et. al., 2018; Hutto & Gilbert, 2014). Depending on the design of the lexical libraries, the sentiment words can generally be classified into either positive or negative polarity and associated with a score for representing the corresponding sentiment strength. These lexical libraries are publicly available to support the text analysis, especially for the evaluation of online review or text contents. The overall sentiment on an issue can be determined by aggregating the sentiment scores of the reviews. However, simply aggregating the sentiment scores may not be sufficient because different reviewers would have different perceptions on the sentiment strengths of sentiment words for expressing their opinions. Different people may use the same word “great” to express their opinions, but the level of sentiment strength is uncertain because the word choice is based on the writer’s subjective judgment. Therefore, this inherent uncertainty has not been sufficiently addressed simply by aggregating the sentiment scores.