An Automatic Event Detection Method for Massive Wireless Access Prediction

The scale of mobile users for parallel access is constrained by the capacity of the base stations. When extremely dense terminal access exceeds the capacity of the base stations, access failure and a performance degradation will occur. The early detection and prediction of important events and the timely detection of possible large-scale terminal access are significant aspects in ensuring the quality of the communication achieved. For the automatic detection of events, methods based on a neural network can learn features automatically without feature engineering and have been proven to be efficient for event detection. As is well known, constructing an adequate input vector that can represent sufficient information is a challenge to a neural network-based approach, particularly for problems caused by Chinese word segmentation and too many unknown communication words. To cope with this problem, a novel representation method that combines the different features with word vectors is proposed to deal with the problem of Chinese event trigger identification. We then use a gated recurrent unit network to train and predict the event trigger and carry out comparative experiments on different methods and feature combinations. The experiment results of the proposed model show that the F1 value can reach 84% for the experimental dataset. Furthermore, the combination of lexical and syntactic features with a neural network was proven to be helpful for this task, although the contributions vary in magnitude for different features. Our study provides directions for further research on the use of lexical and syntactic features with a neural network for an event detection task.