The multi-attribute decision-making method based on online reviews has been widely used in addressing the hotel selection problem. However, due to information overload and the presence of fake reviews, traditional hotel selection methods that rely solely on unverified review analysis can affect the outcome of hotel selections. In this study, a novel three-dimensional helpful review analysis model based multi-attribute decision-making approach for hotel selection is established. Firstly, a new three-dimensional helpful review analysis model that effectively filters out sentiment inconsistency reviews, topic inconsistency reviews, and reviews from invalid sources is proposed. Secondly, this study employs TF-IDF and LDA to extract attributes for hotel selection. We further utilize BERT to extract sentiment level for each attribute. Then, a ranking result for alternative hotels is obtained using a combination compromise solution method (CoCoSo). Finally, we demonstrate its effectiveness and feasibility through a case study of selecting the optimal hotel from TripAdvisor.com.