The following study constitutes an innovative contribution to the Psychotherapy Process Outcome Research which is based on the investigation of the anatomy of a psychotherapy with new Artificial Intelligence tools. Specifically, the research aims to assess whether the scores obtained through the Sentiment Analysis change in the course of twelve clinical sessions and whether there are significant interactions between the actor (patient / therapist), the topics addressed and the scores relating to the emotional polarity (Sentiment Score). Furthermore, the research aims to assess whether the frequency of the identified topics changes longitudinally during the therapy. The clinical sessions were viewed, carefully transcribed and then imported into MATLAB, where the VADER algorithm made it possible to carry out the Sentiment Analysis of the therapeutic sessions. Subsequently, a mixed linear model was implemented and from the analysis of the data emerged that the Sentiment Scores of the therapist and the patient differ significantly: in particular, it is evident that the therapist's expressions are generally more neutral than those of the patient, which are more emotionally intense. Furthermore, the results obtained showed that the Sentiment Scores of some topics (FAM: family, RA: emotional relationship, RS: social relationships) vary significantly between patient and therapist; specifically, the patient's expressions are more positively connoted than those of the psychotherapist. Lastly, following the analysis implemented with the Generalized Additive Mixed Models (GAMMs), it was possible to highlight that the probability of occurrence of some topics (e.g. RA) does not undergo substantial changes at a longitudinal level: this means that some issues are constantly addressed in the twelve clinical sessions, without following an ascending or descending trend; on the other hand, the probability of occurrence of other topics records a decreasing trend (eg. CT, SPS, RT). This result indicates that some issues are progressively addressed to a lesser extent during the therapeutic sessions.
Il presente studio costituisce un contributo innovativo alla Psychotherapy Process Outcome Research, in cui è stata indagata l’anatomia di una psicoterapia breve con i nuovi strumenti di Intelligenza Artificiale. Nello specifico, la ricerca ha come obiettivo quello di valutare se i punteggi ottenuti attraverso la Sentiment Analysis si modifichino nel corso di dodici colloqui clinici e se vi siano interazioni significative tra l’attore (paziente\terapeuta), i topic affrontati e i punteggi relativi alla polarità emotiva (Sentiment Score). Inoltre, la ricerca ha il fine di valutare se la frequenza dei topic individuati si modifichi longitudinalmente nel corso della terapia. I colloqui clinici sono stati visionati, accuratamente trascritti e poi importati in MATLAB, dove l’algoritmo VADER ha consentito di effettuare la Sentiment Analysis delle sedute terapeutiche. Successivamente, è stato implementato un modello lineare misto e dall’analisi dei dati è emerso che i Sentiment Score del terapeuta e della paziente si differenziano significativamente: in particolare, si evince che le espressioni del terapeuta sono generalmente più neutre rispetto a quelle della paziente, le quali risultano più emotivamente intense. Inoltre, dai risultati ottenuti è emerso che i Sentiment Score di alcuni topic (FAM: famiglia, RA: rapporto affettivo, RS: rapporti sociali) variano in modo significativo tra paziente e terapeuta; nello specifico, le espressioni della paziente risultano più connotate positivamente rispetto a quelle dello psicoterapeuta. Infine, in seguito alle analisi implementate con i Generalized Additive Mixed Models (GAMMs), è stato possibile evidenziare che la probabilità di occorrenza di alcuni topic (es.RA) non subisce sostanziali modificazioni a livello longitudinale: ciò significa che alcuni temi vengono affrontati costantemente nei dodici colloqui clinici, senza seguire un trend né ascendente né discendente; d’altra parte, la probabilità di occorrenza di altri topic registra un andamento decrescente (es. CT, SPS, RT). Questo risultato indica che alcuni temi vengono affrontati progressivamente in misura minore nel corso delle sedute terapeutiche.
Nuovi strumenti di Intelligenza Artificiale per la Sentiment Analysis di una psicoterapia breve
DISTASI, GIULIA
2020/2021
Abstract
The following study constitutes an innovative contribution to the Psychotherapy Process Outcome Research which is based on the investigation of the anatomy of a psychotherapy with new Artificial Intelligence tools. Specifically, the research aims to assess whether the scores obtained through the Sentiment Analysis change in the course of twelve clinical sessions and whether there are significant interactions between the actor (patient / therapist), the topics addressed and the scores relating to the emotional polarity (Sentiment Score). Furthermore, the research aims to assess whether the frequency of the identified topics changes longitudinally during the therapy. The clinical sessions were viewed, carefully transcribed and then imported into MATLAB, where the VADER algorithm made it possible to carry out the Sentiment Analysis of the therapeutic sessions. Subsequently, a mixed linear model was implemented and from the analysis of the data emerged that the Sentiment Scores of the therapist and the patient differ significantly: in particular, it is evident that the therapist's expressions are generally more neutral than those of the patient, which are more emotionally intense. Furthermore, the results obtained showed that the Sentiment Scores of some topics (FAM: family, RA: emotional relationship, RS: social relationships) vary significantly between patient and therapist; specifically, the patient's expressions are more positively connoted than those of the psychotherapist. Lastly, following the analysis implemented with the Generalized Additive Mixed Models (GAMMs), it was possible to highlight that the probability of occurrence of some topics (e.g. RA) does not undergo substantial changes at a longitudinal level: this means that some issues are constantly addressed in the twelve clinical sessions, without following an ascending or descending trend; on the other hand, the probability of occurrence of other topics records a decreasing trend (eg. CT, SPS, RT). This result indicates that some issues are progressively addressed to a lesser extent during the therapeutic sessions.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/20.500.14239/1127