Computational Investigation of Phase Transitions in Mental Health

Abstract

Introduction

Several cognitive phenomena can be interpreted through the paradigms of complex and chaotic systems. This enables the paradigms’ mathematical methods to be applied to the data analysis. Some of the most important aspects of human change processes in psychotherapy, e.g., the discontinuous progress, can be explained in terms of their chaotic dynamics [1]. Moreover, the phase transition of self-organizing systems (PT) is characterized by the changes in various dynamical aspects of a patient’s multivariate time series which could computationally be seen e.g., as a shift in the mean–in psychotherapy research defined as “sudden gains or losses” [2].

We are interested in whether the use of machine learning (ML) models will attribute to higher PT detection precision. We will investigate this by (a) comparing our ML models with the Pattern Transition Detection Algorithm [3] (PTDA) which includes various computational algorithms to detect PT and (b) determining the aspects of extending PTDA with ML models to detect PT with higher precision.

Methods

We will use two datasets consisting of participants’ time series of daily self-rating questionnaires and diary entries. The datasets were either gathered from a psychotherapeutic process, or participants’ self-assessments of their depression, anxiety, and stress levels during a month-long study. The features will be extracted from the diary texts with natural language processing (NLP) methods. Then, the dynamical aspects of PT will be computed from the time-series data (e.g., the mean change and periodicity). For (a), ML models will be built, trained, and tested to get the precision scores for the comparison with PTDA. For (b), ML models will be assessed in terms of their convergence utility with PTDA.

Expected Results and Implications

We expect that our ML models will not be able to replace the PTDA as the included algorithms assess various aspects of PT [3]. Moreover, we expect that extending PTDA will attribute to the detection precision. Higher detection precision will enable mental health workers to plan interventions promptly and enable further investigations of PT precursors [2]. This work will give a novel contribution to NLP text features extraction in the mental health domain.

References

[1] G. K. Schiepek et al., “Psychotherapy Is Chaotic—(Not Only) in a Computational World,” Front. Psychol., vol. 8, Apr. 2017.

[2] G. Schiepek et al., “Convergent Validation of Methods for the Identification of Psychotherapeutic Phase Transitions in Time Series of Empirical and Model Systems,” Front. Psychol., vol. 11, Aug. 2020.

[3] K. Viol, H. Schöller, A. Kaiser, C. Fartacek, W. Aichhorn, and G. Schiepek, “Detecting pattern transitions in psychological time series–A validation study on the Pattern Transition Detection Algorithm (PTDA),” PLoS One, vol. 17, no. 3, p. e0265335, Mar. 2022.