This document discusses personalized recommendation models for predicting user ratings of items. It introduces challenges like cold starts and huge datasets. It proposes combining a user's personal interests and social circles by modeling personality, interest similarity between users, and influence between socially connected users. The contribution is a new approach that enforces a user's latent interests in the model. Experiments on extensive datasets show how this approach helps address cold starts. The document outlines recommendation models like matrix factorization and the CircleCon model it proposes.