Online customer reviews generate an electronic word-of-mouth (eWOM) effect in the form of text reviews and ratings, which is an effective way for operators to understand customer experience and satisfaction. Collecting customer data through interviews or questionnaires to explore key factors affecting customer satisfaction is an empirical practice based on constrained time and sample size. However, there are a large number of online customer reviews, and customers can provide detailed feedback on product defects and improvement needs through text without being constrained by questionnaire questions. Therefore, how to effectively mine the dimensional information of customer satisfaction using online comment data and establish a satisfaction analysis framework is a noteworthy issue. This paper develops a two-stage method using text mining and multi-criteria decision making (MCDM) technology. Consider combining unsupervised topic modeling techniques with supervised recurrent neural networks, combining comment text and auxiliary data into the topic model for topic information extraction to obtain satisfaction evaluation attributes. Then, the multi-attribute satisfaction analysis model was extended, which fully considers the qualitative forms of customer judgments and preferences.