In this study, based on the time series characteristics of web browsing records, I proposed a model called Ir-LSTM which draws on the LSTM model, the Skip-Gram embedding method and expand the inputting features with the user information. In addition, a real-time Dynamic Allocation mode is proposed, which is for detecting the traffic burst in real time and correspondingly adjust the correlation coefficient of the model’s output to reach a higher utilization of the server resources and fully maximize the hit ratio of the model.