Abstract: The present disclosure relates to systems and methods to generate user or user group behavior similarities based on computing distances between individual users or user groups and without using demographic data, e.g., based on anonymized data. The methods include creating a multidimensional vector representation for each user in a group by summarizing user's behaviors across various categories. Based on the created vectors, distance calculation and nearest neighbor search are performed to locate users that are most similar to target users. The resulting distance metrics may be used to rank similarities. Additionally, dimensionality reduction may be performed to further distill the behavior information to make the disclosed methods suitable for a variety of analytics and modeling tasks.