KID model realization using memory networks for text-based Q/A analyses and learning

KID (Knowledge-Information-Data) model as a cognitive model was initially proposed for business intelligence by Huang's research team. It aims to mimic human-like cognitive learning by abstracting human information processing into the following three stages: data interpretation, information assimilation and knowledge instantiation. It is a model with highlevel abstraction and a generic framework which can accommodate a variety of realistic cognitive learning applications. 'Memory networks' referring to an architecture for memorizing what learned in networks and performing training jointly with other machine learning methods. It consists of four components: input, generalization, output and response components and longterm memory. In this study, we put efforts on closely looking at both the KID model and components of 'memory networks' architecture so as to understand their relations with which they jointly work for some potential applications. It is found that 'memory networks' can provide a partial realization for the KID model and this paper describes how the 'memory networks', in particular, the enhanced version called 'end-to-end memory' is adopted for realizing the KID model for text-based question answering learning.