Recurrent neural network classifier for Three Layer Conceptual Network and performance evaluation

Abstract

Contextual analysis in dialog is a hard problem. In this paper three layers memory structure is adopted to address the challenge which we refer to as three layer conceptual network (TLCN). This highly efficient network simulates the human brain by episodic memory, discourse memory and ground memory. An extended case structure framework is used to represent the knowledge. The knowledge database is constructed by the collection of target system information and utterances. This knowledge is updated after every dialog conversation. A Recurrent Neural Network classifier is also introduced for classifying the knowledge for the target system. This system prototype is based on doctor-patients dialogs. 78% disease classification accuracy is observed by this system prototype. Disease identification accuracy is depending on number of disease and number of utterances. This performance evaluation is also discussed in details.