Multi-Rate Deep Learning for Temporal Recommendation

Modeling temporal behavior in recommendation systems is an important and challenging problem. Its challenges come from the fact that temporal modeling increases the cost of parameter estimation and inference, while requires large amount of data to reliably learn the model with additional time dimensions. Therefore, it is hard to model temporal behavior in large scale real-world recommendation applications.

In this work, we propose a new deep neural network based architecture that models the combination of user’s long term and short term temporal preferences. We also study the features that are effective for the recommendation applications. For instance, we use rich temporal features for a user from her search logs, and therefore to provide the context for making the recommendation. To train the model efficiently for large scale applications, we propose a novel pre-training method to reduce the number of free parameters significantly to overcome the data sparsity issue. The resulted model is applied to large scale real-world data set from a commercial News recommendation system. We compared to a set of established baselines and the experimental results show that the proposed method outperforms the state-of-the-art significantly.