Neural TV program recommendation based on dynamic long-short term interest

TV program recommendation can help user find interested programs and improve user experience. The heterogeneous information of programs is important for alleviating the problem of data sparsity. In addition, the existing TV program recommendation methods are lacking in dynamics. This paper proposes a neural TV program recommendation based on dynamic long-short term interest (NPR-DLSTI), which mainly includes two modules: program and user encoder. In the program encoder module, we use convolutional neural network and attention mechanism to learn the heterogeneous information of the program and realize program representation. In the user encoder module, we use gated recurrent unit and personalized attention to learn the dynamic change law of users’ interests. Experiments on real data sets show that our method can effectively improve the effectiveness and dynamics of TV program recommendation than other existing models. •We integrate auxiliary information including program title, program labels, program channel and so on, use convolutional neural network (CNN) to mine the semantic association contained in these information, and combine multi-layer attention network to learn program and user encoder representation.•On the basis of using the program textual auxiliary information, we take into account users’ interests can cause dynamic change over time, introduce the time factor related numerical program data. And we combine the deep learning algorithm and the attention network, from the two perspectives of long-term and short-term, learn the dynamic change law of users’ interests.•The effectiveness of our model in learning the change rules of users’ dynamic interests and improving the dynamic nature of the model is verified through the comparative experiments with other recommendation models and the ablation experiments of our model’s own modules.