Exploring the generalizability of discriminant word items and latent topics in online tourist reviews
本研究通过两种分析方法（惩罚支持向量机和潜在狄利克雷分配）探究酒店、餐厅和景点三类旅游企业的正面与负面评论在词汇和主题上的差异，发现判别性词汇的普适性有限，但整体预测准确率未下降。
Purpose Online reviews have been gaining relevance in hospitality and tourism management and represent an important research avenue for academia. This study aims to illustrate the discrimination between positive and negative reviews based on single word items and the sector-specific relevance of hidden topics. Design/methodology/approach By probing two parallel approaches of entirely unrelated analytical methods (penalized support vector machines and Latent Dirichlet Allocation), the analysts explore differences in language between favorable and unfavorable reviews in three service settings (hotels, restaurants and attractions). Findings The percentage of correctly predicted positive and negative review reports by means of individual word items does not decrease if reports from the three tourism businesses are analyzed together. Originality/value However, there is limited generalizability of the discriminant words across the three businesses. Also, the latent topics relevant for generating customers’ review reports differ significantly between the three sectors of tourism businesses.