Recommender systems provide personalized content from various choices by mining users’ past preferences. Recommendation helps to overcome the information overload problem as the available alternatives consume a large amount of data. In some applications, a group of people are involved in the process of generating a recommendation. This paper focuses on “automatically detected groups” formation for order and flexible preferences in group recommendation using locality-sensitive hashing. The MinHash technique is applied on a characteristic matrix to generate the signature matrix. The signature matrix is the reduced representation of the characteristic matrix and preserves the Jaccard similarity to a great extent. Locality-sensitive hashing is applied on the signature matrix to determine similar users efficiently. Similar users will be the members of an automatically identified group. Therefore, the group members are maximally satisfied with recommended items. This work also studies the performance of benchmark clustering approaches in group formation. We experimented on real-world datasets and found that the proposed models to identify communities in group recommendation maximizes consensus among users in a group.