Towards a Computational Framework for Psychological Formulation: Conceptual Principles

Case formulation plays a central role in psychological therapy, yet its narrative form limits empirical analysis and digital integration. This paper develops a theoretical framework that treats formulation as a structured representation of causal cognition—a generative process linking clinical observations to explanatory hypotheses. Drawing on advances in large language models (LLMs) and graph-based methods, it outlines a conceptual workflow for converting clinical text into machine-readable causal graphs constrained by psychological ontologies. The paper is conceptual in nature and employs simple worked examples to demonstrate how LLMs can identify causal information within narrative formulations and convert it into structured graphs. Within this framework, deep-learning semantic-similarity methods and graph-comparison metrics such as the Jaccard index—implemented in R—serve as illustrative demonstrations of how researchers might begin exploring causal representations using accessible tools, with more advanced graph-learning techniques offering promising directions for future work. Given the stochastic nature of LLMs and the evolving character of clinical reasoning, the aim is not traditional reliability testing but the development of representational principles that approximate clinicians’ explanatory models. Future systems may implement this interactively, prompting clinicians to confirm whether their reasoning has been accurately represented diagrammatically, thereby maintaining alignment between human judgement and its computational depiction. The paper’s contribution is therefore conceptual rather than empirical, offering foundational representational principles and a roadmap for integrating LLMs, graph analytics, and visual interfaces to support clinical reasoning, digital formulation tools, and the development of computational theories of psychological formulation.