The present work is part of the Process-Outcome Research, which examines the processes and results leading to a positive outcome of psychotherapy (Tompkins & Swift, 2014). In recent years, we have seen the gradual implementation of artificial intelligence and, in particular, technologies such as Text Mining and Machine Learning (Atkins et al., 2012). This study aims to verify the possibility of identifying both new and significant topics for a given psychotherapy, observing their trend and change over time, through a quantitative and qualitative evaluation of the weight of the words in their context. The psychotherapy analyzed consists of 26 interviews conducted with a cognitive-neuropsychological approach and transcribed following Giorgi's (1985) phenomenological-descriptive analysis methodology. Then, the entire corpus was divided into three different phases (initial, central and final) to obtain the temporal sequence. At first, we performed a quantitative analysis using the Latent Dirichlet Allocation algorithm (LDA), which identified eight topics for each phase. Then, we labelled each topic through qualitative analysis. Finally, we compared the assigned labels with the nuclear topics previously identified by the experienced therapists on the basis of their own knowledge. During the reading of the transcripts, we noticed an existential uneasiness referred to three different topics: affective relationship (RA), family (FAM) and occupation-study (OS). The analyses carried out showed the presence of RA, FAM and OS topics co-occurring with suffering/pathology/symptoms (SPS) in the early stages of therapy. In the final stage, RA was identified as a nuclear theme, OS was associated with planning (PRG), while FAM was not detected. We identified two new topics not previously considered by experienced therapists: introspection and interpersonal relationships. In addition, specific topics emerged from the analyzed therapy. By looking at the results, we can see that psychotherapy has led to a positive change in the patient. Despite the limitations of this study, it was possible to observe the benefits that Text Mining may bring to the Process-Outcome Research.
Il presente lavoro si inserisce nell’ambito della Process-Outcome Research, la quale si occupa dello studio dei processi e dell’outcome che portano ad un esito positivo della psicoterapia (Tompkins & Swift, 2014). Negli ultimi anni, in quest’ambito, si assiste alla progressiva implementazione dell’intelligenza artificiale ed in particolar modo di tecnologie come Text Mining e Machine Learning (Atkins et al., 2012). Questo studio si pone l’obiettivo di verificare la possibilità dell’identificazione di nuovi topic e topic significativi per una data psicoterapia, osservando il loro andamento e cambiamento nel tempo, attraverso una valutazione quantitativa e qualitativa del peso delle parole nel loro contesto. La psicoterapia analizzata è costituita da 26 colloqui condotti con un approccio cognitivo-neuropsicologico, trascritta seguendo la metodologia dell’analisi fenomenologico-descrittiva di Giorgi (1985) e successivamente divisa in tre fasi (iniziale, centrale e conclusiva) per ottenere la scansione temporale. Il primo passaggio è stato quello di eseguire un’analisi quantitativa tramite l’implementazione dell’algoritmo Latent Dirichlet Allocation (LDA), il quale ha consentito l’individuazione di 8 topic per ognuna delle tre fasi. Successivamente, tramite analisi qualitativa, gli sperimentatori hanno etichettato ciascun topic e in seguito confrontati con quelli nucleari identificati precedentemente dai terapeuti esperti sulla base della loro esperienza clinica e psicoterapeutica. Dalla lettura della terapia è emerso un disagio esistenziale orientato su tre topic: Rapporto Affettivo (RA), Famiglia (FAM) e Occupazione-Studio (OS). Dalle analisi condotte è emersa la presenza dei topic RA, FAM e OS in co-occorrenza con il tema Sofferenza/Patologia/Sintomi (SPS) durante le prime due fasi della terapia. Nel blocco conclusivo, invece, RA è stato individuato come topic nucleare, il tema OS in correlazione alla progettualità (PRG) mentre FAM non è stato rilevato. Sono stati identificati due nuovi topic non considerati precedentemente dai clinici esperti: introspezione e relazioni interpersonali. Inoltre, sono emersi topic specifici per la terapia analizzata. Dai risultati ottenuti si può ritenere che la psicoterapia abbia indotto nella paziente un cambiamento positivo. Nonostante i limiti presenti in questo studio è stato possibile osservare i vantaggi che la Text Analysis possa apportare alla Process-Outcome Research.
Topic Modeling e Process-Outcome Research: il contributo dell'intelligenza artificiale all'analisi di una psicoterapia conclusa
AIESI, ELEONORA
2021/2022
Abstract
The present work is part of the Process-Outcome Research, which examines the processes and results leading to a positive outcome of psychotherapy (Tompkins & Swift, 2014). In recent years, we have seen the gradual implementation of artificial intelligence and, in particular, technologies such as Text Mining and Machine Learning (Atkins et al., 2012). This study aims to verify the possibility of identifying both new and significant topics for a given psychotherapy, observing their trend and change over time, through a quantitative and qualitative evaluation of the weight of the words in their context. The psychotherapy analyzed consists of 26 interviews conducted with a cognitive-neuropsychological approach and transcribed following Giorgi's (1985) phenomenological-descriptive analysis methodology. Then, the entire corpus was divided into three different phases (initial, central and final) to obtain the temporal sequence. At first, we performed a quantitative analysis using the Latent Dirichlet Allocation algorithm (LDA), which identified eight topics for each phase. Then, we labelled each topic through qualitative analysis. Finally, we compared the assigned labels with the nuclear topics previously identified by the experienced therapists on the basis of their own knowledge. During the reading of the transcripts, we noticed an existential uneasiness referred to three different topics: affective relationship (RA), family (FAM) and occupation-study (OS). The analyses carried out showed the presence of RA, FAM and OS topics co-occurring with suffering/pathology/symptoms (SPS) in the early stages of therapy. In the final stage, RA was identified as a nuclear theme, OS was associated with planning (PRG), while FAM was not detected. We identified two new topics not previously considered by experienced therapists: introspection and interpersonal relationships. In addition, specific topics emerged from the analyzed therapy. By looking at the results, we can see that psychotherapy has led to a positive change in the patient. Despite the limitations of this study, it was possible to observe the benefits that Text Mining may bring to the Process-Outcome Research.È consentito all'utente scaricare e condividere i documenti disponibili a testo pieno in UNITESI UNIPV nel rispetto della licenza Creative Commons del tipo CC BY NC ND.
Per maggiori informazioni e per verifiche sull'eventuale disponibilità del file scrivere a: unitesi@unipv.it.
https://hdl.handle.net/20.500.14239/2866