As the most growing sector for Indonesia's internet economy during the last five years, e-commerce generates online customer reviews that can be a source for information and giving hints for potential improvements for various stakeholders. Online reviews consist of review text and rating, each of which describes customer's concern and satisfaction in purchasing items online. However, online review text is unstructured, and its relation with rating is hardly observed. This study examines 132,085 online reviews about Xiaomi mobile phones on three major e-commerce websites in Indonesia: Shopee, Bukalapak, and Blibli by text mining and quantitative modeling to correlate reviews with ratings. Online reviews are classified into eight distinct topics, and the relationships between each topic and review rating are analyzed. Multilinear regression is implemented to examine the valence and strength in the relationship between each topic-rating. The result shows that there are more topics with a negative relationship with rating, with several topic differences between the three websites. Further improvements should be focused on the most impactful topics, which are referred to the mobile phone features, such as CPU & hardware, system, and physical appearance. After-sales service is also concerned at Bukalapak and Blibli. These relationships are explained further by the valence expressed by customers in review texts. The implications of this study can be applied for academic purposes, e-commerce companies, customers, sellers, and mobile phone companies.