In order to address the challenges of high concealment, diverse forms, and scarce annotations in tourism-related online reviews, a multi-dimensional credibility evaluation framework was used to investigate the quantification of review credibility. The framework integrates content, semantic, and behavioral features to constrain the space for deceptive reviews. An empirical study was conducted based on real reviews from Harbin Ice and Snow World. A rule-based filtering layer was developed to remove low-camouflage noise. A semantic deviation metric was constructed using noun-level BERT embeddings and adaptive DBSCAN. A TF-IDF-weighted similarity graph with Louvain community detection was built to identify duplicate clusters. Fine-grained sentiment-score inconsistency was measured via RoBERTa. The results show that the proposed framework effectively suppresses deceptive reviews and provides an interpretable, deployable solution for platform governance. It is concluded that the multi-dimensional credibility evaluation framework offers an effective tool for assessing the credibility of online tourism reviews without relying on manual annotation.