Machine Learning and Human Cognition Combined to Enhance Knowledge Discovery Fidelity

The objective of this work is knowledge discovery in large-scale audio files by performing a Cognitive Analysis – CA –, where the knowledge is extracted from transcribed customer service conversations taking into consideration individual cognitive styles to mimic the human cognitive process and maximize the correct meaning interpretation information in a given context. We make the following three contributions: (i) integrate a Cyber Cognitive Identity model – CCI – that states the cognitive profile an individual has for interacting in cyberspace, which yields superior fidelity to identify the meaning of spoken sentences following Sternberg's Thinking Style Inventory (TSI). In particular it guides an analysis grounded in peers' cognitive styles to index words by dimension; (ii) a novel method that extends the Latent Dirichlet Allocation (LDA) approach to a multidimensional partially supervised machine learning model with the help of the psychological activation theory Adaptive Control of Thought – ACT; (iii) an improvement of the Exploratory Data Analysis – EDA–suggested by De Mast and Trip, envisioned as an extended approach to obtain high-fidelity data where topics of a three-dimensional corpus are clustered according to cognitive categorizations. Using speech-to-text software, we transcribed and evaluated 27 500 calls from 206 German-speaking teleworkers combining these three complementary methods and achieved significant fidelity to generate a hypothesis based on individuals' cognitive affinities.