In E-learning context, we can recommend pedagogical re- sources to help learners. In this context, the recommender proposes the nearest resource(s) in term of similarity, where the scarcity of resources may a↵ects seriously the quality of predictions. To make accurate predic- tions we begin in determining the scarce resources in order to be taken into account with the recommendation process. To acheive this objective we use the unsupervised neural network I2GNG (Improved Incremental Growing Neural Gas).