Abstract
Content-based recommender systems (CBRS) are tools that help users to choose items when they face a huge amount of options, recommending items that better fit the user's profile. In such a process, it is very interesting to know which features of the items are more important for each user, thus the CBRS provides them higher weight. The Term Frequency-Inverse Document Frequency (TF-IDF) method is one of the most used for weighting of features, however, it does not provide the best results when the features are multi-valued. In this contribution, it is proposed a new method for obtaining the weights of the features by means of entropy and coefficients of dependency.