This article presents 3 weaknesses for Collaborative Filterings (CF) algorithms. These are: data sparsity (difficult to calculate user similarity because of very few commonly rated items), cold-start users (new users with no ratings – or no friends in our case) and malicious users attacking Recommender Systems (e.g. by copy-profile attack). They conduct experiments using a large (real) database from a website and prove that having a trust metric between users is much more accurate than common CF algorithms. We do not necessarily have to include a trust value for every relationship in Shopster but we can incorporate trust as a measure of how trustworthy a user is, based on their divergence from actual average ratings but also on how helpful their posts and comments (if any) are rated by other users.
Source: Trust-aware Recommender Systems