In the current study, was investigated the contribution that Artificial Intelligence (AI) can offer to Psychotherapy Process-Outcome Research. In particular, Topic Modeling was adopted as a method for the text analysis of a brief psychotherapy, conducted using the approach of Psicoterapia Cognitivo Neuropsicologica. The interviews were first analyzed and labeled with Topic by expert raters, following a procedure similar to Giorgi's (1985) descriptive phenomenological method. Subsequently, the automated analysis was performed with the Latent Dirilecht Association (LDA) algorithm, a probabilistic model capable of identifying clusters of words with semantic similarities (topics) that better describe a document. The first aim of this work was to investigate whether the performance (o pattern) of the Topics, identified with qualitative method, followed a very specific anatomy of the therapeutic structure. The second purpose was to evaluate how AI could offer its contribution to the discrimination of topics. Through an initial statistical analysis with the Generalized Additive Mixed Models (GAMMs), it turned out that the Topics "Macro-contesto esistenziale" (MCE), "Relazione Affettiva" (RA) and "Sofferenza/Patologia/Sintomi" (SPS) followed a logical probability occurrence over time, according to a strategically oriented clinical action. Furthermore, the application of the LDA has shown how this text mining model is able to detect the same topics identified by the expert raters. Besides, this tool was also capable of suggesting more specific topics and new topics. Although some limitations have arisen, mainly due to the lack of literature inherent to these issues, the results obtained are encouraging. In fact, refining the adopted methods, it will be possible to obtain even better results in future studies.
Nello studio corrente è stato indagato il contributo che l’Intelligenza Artificiale (IA) può offrire alla Psychoterapy Process-Outcome Research. In particolare, è stato adottato il Topic Modeling come metodo per l’analisi testuale di una psicoterapia breve, condotta secondo l’approccio Cognitivo Neuropsicologico. I colloqui sono stati, dapprima analizzati ed etichettati con i Topic da valutatori esperti, seguendo un metodo simile a quello fenomenologico descrittivo di Giorgi (1985). Successivamente, è stata effettuata l’analisi automatizzata con l’algoritmo Latent Dirilecht Association (LDA), un modello probabilistico in grado di individuare gruppi di parole (i topics) che meglio rappresentano un documento. Il primo obiettivo del presente lavoro è stato indagare se l’andamento dei Topic, individuati qualitativamente, seguisse un’anatomia ben precisa della struttura terapeutica. Il secondo obiettivo è stato quello di valutare in che modo l’IA potesse offrire il suo contributo alla discriminazione di topics. Tramite una prima analisi statistica con i Generalized Additive Mixed Models (GAMMs), è emerso come i Topic “Macro-contesto esistenziale” (MCE), “Rapporto affettivo” (RA) e “Sofferenza/Patologia/Sintomi” (SPS) seguissero una probabilità di occorrenza sensata nel corso del tempo, secondo un agire clinico strategicamente orientato. Inoltre, dall’applicazione di LDA è stato evidenziato come questa tecnica di analisi testuale sia in grado di individuare gli stessi topics individuati dai valutatori esperti. Non solo, questo strumento è stato anche in grado di suggerire topics più specifici e nuovi topics. Nonostante siano emersi alcuni limiti, dovuti soprattutto alla carenza di letteratura inerente a queste tematiche, i risultati ottenuti sono di buon auspicio. Infatti, perfezionando le metodologie adottate sarà possibile negli studi futuri ottenere risultati ancora migliori.
Il contributo dell'intelligenza artificiale alla process-outcome research in psicoterapia: il "Topic Modeling" adottato per l'analisi testuale di una psicoterapia breve
GRADASCHI, MARCO
2020/2021
Abstract
In the current study, was investigated the contribution that Artificial Intelligence (AI) can offer to Psychotherapy Process-Outcome Research. In particular, Topic Modeling was adopted as a method for the text analysis of a brief psychotherapy, conducted using the approach of Psicoterapia Cognitivo Neuropsicologica. The interviews were first analyzed and labeled with Topic by expert raters, following a procedure similar to Giorgi's (1985) descriptive phenomenological method. Subsequently, the automated analysis was performed with the Latent Dirilecht Association (LDA) algorithm, a probabilistic model capable of identifying clusters of words with semantic similarities (topics) that better describe a document. The first aim of this work was to investigate whether the performance (o pattern) of the Topics, identified with qualitative method, followed a very specific anatomy of the therapeutic structure. The second purpose was to evaluate how AI could offer its contribution to the discrimination of topics. Through an initial statistical analysis with the Generalized Additive Mixed Models (GAMMs), it turned out that the Topics "Macro-contesto esistenziale" (MCE), "Relazione Affettiva" (RA) and "Sofferenza/Patologia/Sintomi" (SPS) followed a logical probability occurrence over time, according to a strategically oriented clinical action. Furthermore, the application of the LDA has shown how this text mining model is able to detect the same topics identified by the expert raters. Besides, this tool was also capable of suggesting more specific topics and new topics. Although some limitations have arisen, mainly due to the lack of literature inherent to these issues, the results obtained are encouraging. In fact, refining the adopted methods, it will be possible to obtain even better results in future studies.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/20.500.14239/1935