Popular recommendation algorithm like collaborative filtering by matrix factorization takes into account the accounts for user and item biases (the simplest method is to normalize the ratings of a particular user by the average of ratings of that user).