An Explainable TV Program Recommendation Model Based on Attention Mechanisms

Abstract

TV program recommendation can help users find programs from a large amount program information that is often overwhelming and confusing. The recommendation systems based on neural network architectures have been successfully applied to TV program recommendation. However, owing to the “black box” of neural network architectures, it is hard to explain the methodologies behind these recommendations and this often leads to hesitancy of users in using recommendation. Here we propose a label-based model with hierarchical attention mechanisms for TV program recommendations. We introduce the attention in two dimensions of “user” and “program” to measure the importance of labels in TV programs. In the process of training the model using real world data, we can visualize the attention to explain the changes in the model. For the test data, we use the attention to interpret the recommended samples. Experiments demonstrate exceptional interpretability of our model.