A Semantic Parsing Based LSTM Model for Intrusion Detection

Nowadays, with the great success of deep learning technology, using deep learning method to solve information security issues has become a study hot spot. Although some literal works have tried to solve intrusion detection problem via recurrent neural network, these methods do not give a detailed framework and specific data processing progress. We propose a novel semantic parsing based Long Short-Term Memory (LSTM) network framework in this paper. The proposed method uses the semantic representations of network data. The novel conversion process of various forms of network data to semantic description is given in detail. Experiments on NSL_KDD data sets show our proposed model outperforms most of the standard classifier. Results show that the semantic description has reserved information of the data and our semantic parsing based LSTM model provides a novel way to solve anomaly detection.