Abstract: A set of customers’ reviews about restaurants has been analyzed syntactically and semantically for deducing syntactic, contextual and semantic features to leverage the textual similarity metrics. In this study an approach for rule based extracting semantic features from customer’s reviews have been proposed. The features were extracted based on the external knowledge base (Word Net), co-occurrence and distributional similarity among the reviews’ aspects and descriptors and then an algorithm has been created for grouping the aspects naturally by basing on the computed similarity features. The proposed system has applied on the Yelp academic challenges dataset and the results have shown encouraged performance.