The vast body of reviews for individual items makes it difficult for users to extract useful information. And the opinions expressed by these users can be easily influenced through the communication services provided by the E-Commerce systems. A number of works are proposed for information extraction from a large review corpora, and these techniques may release users from the tiresome task of reading reviews. However, such extracted summaries are lack of immediacy, and could not keep the reviews' narrative structures. Aiming at enhancing the diversity of reviews and eliminating the above methods' defects, we propose a local community based algorithm to group reviewers and recommend the original reviews to users. Our method utilizes similarity-based sparsification techniques to identify the edge types that connected two nodes to determine these two nodes are in the same community or not. Since such identification procedure only evolves the neighbors of the target nodes, it can be set on the client side, and can be accomplished efficiently. We conduct comprehensive experiments to demonstrate the accuracy of our algorithm, and provide the discussions and explanations about the phenomena appeared in the experimental results.