Use of Natural Language Processing to Discover Evidence of Systems Thinking

In prior work, we proposed elements of a generalized core language for systems thinking that characterizes a systems approach. We developed this foundational language through the relationship between critical thought and systems in the definition of systems thinking. We submitted our interpretation of this language, i.e., the Dimensions of Systems Thinking (DST), to the systems community for further development. Assuming that the DST provides a basis in language to express systems thought in a systems design or analysis, our objective in this phase is to present a strategy for discovering objective statistical evidence of systems thinking based on the linguistic elements of the DST in an unread corpus of documents. With that goal, we consider statistical semantic characterization of systems thought through a process of supervised learning, term frequency and inverse document frequency (tf-idf), cosine similarity, and naïve Bayes classifiers, specifically Rocchio classifiers and quadratic discriminant classification. Finding evidence of systems thinking in unread text through natural language processing establishes a foundation for computationally assessing systems thinking quality in a document or corpus of design-related documents without having to read the subject document or corpus. Once demonstrated, the method can be applied to systems thinking fluency in design-focused documents across a range of application areas, such as transportation, healthcare, and environmental policy. This study demonstrates that capability with a high degree of selectivity. Our approach establishes correlation and allows us to assess the systems thinking quality of a corpus of unread analysis and design studies on life cycle assessment. We show that assigning a prior probability improves that assessment. We posit that quantitative relations between the specific dimensions of systems thinking and a document would be useful for recognizing good systems thinking and improving the quality of systems analyses.