Study on Sparsity of Recommender System in University Library

Due to the lack of readers’ rating data, university library recommender system is facing the problem of sparsity of data. This study proposes a general logarithmic transformation model which can convert the reader’s implicit feedback data to the rating, thus alleviating the sparsity to a certain extent. Because logarithmic transformation can use different base, several logarithmic transformation methods are analyzed and compared from different angles. The model-based collaborative filtering is used to compare the recall and accuracy of these methods to make full use of the technical advantage based on matrix factorization to further alleviate the data sparsity problem. The experimental results show that the proposed general logarithmic transformation model can play the role of modifying the data skew, compressing the variable scale, reducing the value of the calculation and so on, and the results of the model are interpretable. Moreover, when the suitable k value is selected, the recommended results of different logarithmic transformation methods can approach the optimal solution in a finite experiment.