Abstract:Recently, video contents application (e.g. multimedia on-demand service, short video playback, etc.) are provided by more and more automobile manufacturers in vehicular networks. However, the huge traffic poses a large burden on the limited wireless network resources. Edge caching technology is a promising solution, where base stations act as edge cache nodes. There are limitations that base stations have restricted coverage and the cost of deployment is a concern. Unmanned Aerial Vehicles(UAVs) has been widely used due to its lightweight and ease of deployment. Each vehicular user requests video contents of interest from the content provider and nearby UAVs which precache a certain video contents in advance deliver video contents to the requester, i.e., vehicular users. However, the storage capacity of UAVs is limited. In order to efficiently utilize the storage resources of UAVs, it is necessary for UAVs to proactively cache video contents with higher popularity. It is a challenge that the popularity of video contents are unknown and time-variant. Thus, we need to predict the popularity of video contents. In this paper, a prediction scheme based on LSTM-Attention method is proposed. Specifically, historical requests of video contents are used to make a prediction on the request of video contents in the next time step in the LSTM model. Attention mechanism is added to improve the accuracy of the prediction of the popularity of video contents. The real-world dataset is used to demonstrate that the proposed LSTM-Attention prediction scheme can enhance the prediction performance of popularity.

LSTM-Attention-Based Video Popularity in UAV-Assisted Vehicular Networks