Preferential Attachment, Aging and Weights in Recommendation Systems

In the present work, algorithms based on complex network theory are applied to Recommendation Systems in order to improve their quality of predictions. We show how some networks are grown under the influence of trendy forces, and how this can be used to enhance the results of a recommendation system, i.e. increase their percentage of right predictions. After defining a base algorithm, we create recommendation networks which are based on a histogram of user ratings, using therefore an underlying principle of preferential attachment. We show the influence of data aging in the prediction of user habits and how the exact moment of the prediction influences the recommendation. Finally, we design weighted networks that take into account the age of the information used to generate the links. In this way, we obtain a better approximation to evaluate the users’ tastes.