Systems and methods for determining software recommendations for a user. A first application list of applications installed on a user's computer is received. A distribution score is determined for each application in the first application list. A set of least distributed applications is determined based on the distribution score. A similarity score is determined for each user in a set of users having one or more applications of the set of least distributed applications installed on their respective systems. A second list of applications is determined based on applications installed by users in the set of users having a similarity score above a threshold. Recommendations for applications in the first list of applications are determined based, at least in part, on typicality scores for the applications.