Deep Tag Recommendation Based on Discrete Tensor Factorization

In the recent years, tag recommendation is becoming more and more popular in both academic and industrial community. Although existing models have obtained great success in terms of enhancing the performance, an important problem has been ignored - the efficiency. To bridge this gap, in this paper, we design a novel discrete tensor factorization model (DTF) to encode user, item, tag into a unified hamming space for fast recommendations. More specifically, we first design a base model to translate the traditional pair-wise interaction tensor factorization (PITF) into its discrete version. Then, to provide our model with the ability to involve content information, we further extend the base model by introducing a deep content extractor for more comprehensive user/item profiling. Extensive experiments on two real-world data sets demonstrate that our model can greatly enhance the efficiency without sacrificing much effectiveness.