General-purpose sensor message parser using recurrent neural networks with stack memory

The growth in the usage of the Internet of Things (IoT) has resulted in the deployment of diverse networks. However, the multiple networking interfaces and embedded protocols pose a significant challenge to commu-nication compatibility. To tackle this problem and establish a flexible networking framework, we propose the implementation of a general-purpose message parser utilizing a recurrent neural network model with stack memory (RNN-SM). This parser has the ability to extract crucial keywords from the various communication network messages, which are trained on multiple network protocol specifications. During the training phase, the RNN-SM predicts candidate keywords and cross-references them with predefined keywords in an expandable dictionary, thus improving the accuracy of keyword extraction. Additionally, we have introduced the concept of minimum prediction fork level as a hyperparameter to balance the simplicity and flexibility of the RNN-SM. The proposed parser proves to be an effective solution in facilitating smooth communication between multiple de-vices and also has the added benefit of filtering out noise. The RNN-SM's robust keyword extraction capability holds up even in noisy environments, making it a reliable solution for the compatibility challenges posed by the IoT.