Deep Learning for Latent Events Forecasting in Content Caching Networks

A novel Twitter context aided content caching (TAC) framework is proposed for enhancing the efficiency by taking advantage of legibility and massive volume data. For purpose promoting efficiency, three machine learning models are to predict latent events popularity, utilizing collected data with geo-tags geographic information adjacent base stations (BSs). Firstly, we propose a Dirichlet allocation (LDA) model forecasting because superiority LDA in natural language processing (NLP). Then, conceive long short-term memory (LSTM) skip-gram embedding approach LSTM continuous skip-gram-Geo-aware popularity forecasting. Furthermore, associate strategy. Lastly, non-orthogonal multiple access (NOMA) based transmission scheme. Extensive practical experiments demonstrate that: 1) TAC outperforms conventional capable being employed applications thanks associating ability public interests; 2) conserves (NLP) data; 3) perplexity lower compared approach; 4) evaluation demonstrates that hit rates tweets vary from 50% 65% rate contents up approximately 75% smaller space algorithms. Simulation results also shows NOMA-enabled scheme least frequently used (LFU) 25%.