RECOMMENDATIONS BASED ON RELEVANT FRIEND BEHAVIORS

摘要

Embodiments are directed towards determining dependent interest affinity values between users to identify users that may mirror interests and thereby have an increased probability of becoming friends. A plurality of tracked online activities are classified into a plurality of interests categories, and used to determine weighted scores for each interest based on a quantity and quality of related activities for the interest. A proportional score for each interest is also determined and used with the weighed scores to generate dependent interest affinities between pairs of users. Interest indices are obtained and rank ordered for a given user and another user based on relevant dependent interest affinities. The resulting interest indices may be filtered based on a variety of criteria. At least some information about the related other users may be displayed to the given user based on the rank ordering, as possible mirrored friends.