A system and method of predicting, for a user selected from a group of users, the reactions of the selected user to items sampled by one or more users in the group but not sampled by the selected user. The predictions are based on other items previously sampled by that user. A scalar rating is defined for each item sampled by the selected user to represent the reaction of the selected user to that item. The selected user is successively paired with other users in the group who have defined scalar ratings for at least some of the items sampled by the selected user to determine the overall difference in ratings for items sampled by both members of each successive pair. One or more of the other users are designated as predicting users and a weighting value is assigned to each of the predicting users based on the overall difference in ratings between that predicting user and the selected user. The weighting values are applied for each item unsampled by the selected user to proportionally alter the difference between a rating previously predicted for each identified item and any actual ratings of that item by the predicting users to adjust the reaction predictions for the selected user.