Quantifying Relationships in Fuzzy Cognitive Maps Based on Content Analysis of Unstructured Research Texts

FCM projects often rely on knowledge-based approaches, such as expert interviews, which can be challenging to conduct because they require extensive expert participation. We propose a novel alternative to create FCM models based on expert knowledge that is already codified in research texts and other publications. In this approach, thematic network analysis identifies FCM concepts and signed causal connections, while computation of t-coefficient is used to determine the weights of the identified edges. We introduce and evaluate the approach in the context of a real-world system modeling project that is based on 47 carefully selected, peer reviewed research publications on Theory on High Reliability Organizations. We evaluate the approach and the FCM model based on three strategies: (1) comparing edge weights, assigned with t-coefficient, against the commonly used-coefficient and cosine similarity index, (2) comparing the behavior of the FCM model that results from this work against known system behavior, and (3) comparing FCM model behavior against the system knowledge of safety experts in the offshore oil and gas industry. The work shows that this first application of t-coefficient in FCM modeling provides a practical, robust, and sufficiently sensitive means to extracting causal weights from manually coded text.