Toward the automation of family semantic grids: a pilot exploration of text mining for evidence-informed systemic-relational psychotherapy

Introduction: Retracing the theme of what can legitimately be considered “evidence” in psychotherapy, the present work shows the application of an ad hoc, rule-based text mining program specifically designed to analyze therapy session transcripts. Developed within a systemic-relational framework, and grounded in Semantic Polarities Theory, the program works through a hybrid human-in-the-loop approach and provides data for subsequent descriptive and inferential analyses on meaning transformation starting from consultation sessions. Participants: The present study examines the clinical case of an adolescent girl with untreated past non-suicidal self-injury (NSSI) and suicidal ideation. Before moving to the joint family work phase, family therapy consultation was conducted in separate sessions within an alternating parent-adolescent format. Methods: Overall semantic activity was compared across events with Wilcoxon rank-sum tests. In categorical comparisons, data were analyzed using chi-square tests, with df-scaled Cramer’s V to measure effect sizes and Monte Carlo p-values. Initial subjects’ positioning profile across Family Semantic Grids (FSGs) was described using semantic heatmaps and main polarities use. Linear Mixed-Effects (LME) models were implemented to track longitudinal changes in the subjects’ positioning, specifically investigating the therapist’s influence. Results: Results indicated that the therapist introduced a significantly lower proportion of new meanings across all comparisons during the consultation phase. LME models showed significant intensity shifts toward positive-valued meanings, supported by ICC, R2m and R2c values, and inform on the significant specific use of Narrated Semantic Polarities (NSP) and Interactive Semantic Polarities (ISP) across participants. Discussion: Overall, the study supports the feasibility of a semi-automated process for coding and for subsequent analyses of family semantic polarities. In fact, performed analyses demonstrate how therapist positioning and meaning management through re-expression and semantic shifts significantly influenced family members’ change in positioning, an effect especially evident for the identified patient. In conclusion, rule-based text mining can yield process indicators that remain clinically legible. Read as “semantography,” results provide a compact, data-grounded map of coupling and early change while staying parsimonious on case disclosure. Limits and future directions of the present pilot study are discussed.