Outcome Related Factors In Counseling Process

It has been decades since we first start to research on what process and outcome of psychotherapy,within these years,a large body of results was accumulated.Those result helped us to have a better understanding of process and outcome,and mechanism of psychotherapy.However,more and more researchers have found out that even though we have calculated vast knowledge,those knowledges have not produced as much progress in psychotherapy services.One of the reasons might be the gap between research and practice(Goldfried,2019).To closing the gap between research and practice,research on the text that generated from the psychotherapy sessions might help.However,due to the limitation of the methodology,up to now,it is still hard and could reach little consensus to directly use the text data in research.This study was aimed at using some new technology-based method to overcome the limitation on the psychotherapy transcription-based research,and to help to close the gap between research and practice.Specifically,we have used a qualitative research method,qualitative comparative analysis method,classifier that based on Language Inquiry and Word Counts dictionary,and Topic Model to analyze the text of psychotherapy process,especially emphasis on therapist words in session.Two layers of information was obtained by doing this research,namely 1)are those methods fit the condition of transcription-based research and could generate meaningful result;2)what kind of event that happened in session could help to reach a better outcome in session.Research one was asked the experienced therapists to label out the high-and low-quality reply of therapists in the research material,by using inductive method,the criteria of experienced therapists'evaluation was summarized.It has been found that firstly,more experienced therapists do produce more high-quality responses,however,it was the same group who produced most low-quality response which keep changing the subject,too.So,experience could be the double blade sword.Secondly,those response that focused on emotion more tends to be high quality,and those which focused on problems was rarely be seen as high quality.Finally,blame the client,swift reply without fully understands clients'point,and put hardly answered questions to client was the common mistakes that therapists made.Research two employed a set-theory-based method,fuzzy-set qualitative comparative analysis(fs/QCA),to explore what factors could help to build a better working alliance.QCA was a set theory-based method that employed Boolean logic.By adopting the concept of degree of membership and the coverage rate,unique feature of multiple cases was explored.Compared to the traditional method,this method was more in to explore the asymmetric causality,therefore,this could be helpful to clear up the process and outcome issue in psychotherapy.The result indicates an interaction between therapists'performance and client's symptom severity which have impact on the working alliance.In the four conditional assembly that lead to better alliance,three out of four were with the condition of clients'lower severity,and the only combination with a higher severity was accompanied by therapists' high-quality performance,while clients' less arousal after sessions.And if therapist keeping making mistakes in session without highlight performance,then the alliance stays low.Research three was use the Textmind data set as the session index to explore the possibility of using classification algorithm to tell the difference between therapy session with different outcomes.We have used two supervised algorithm which were liner discrimination model and naive bayes model,and two unsupervised algorithm which were principal component analysis and t-distributed stochastic neighbor embedding,to mining the Textmind data of sessions.The result revealed that the supervised method tends to have a good effect on classing the data,but not yet reach the level of direct using in practice.However,the result indicated that there is potential application of using this result to predict the outcome of the session.Unsupervised method was another story.The result of PCA has almost no consistent with the outcome data,while the t-SNE model indicated that Textmind index of different outcome sessions do clustering together,but in different sub-group,which might indicate that even the same outcome sessions might be grouped as different sub-group.Research four was adopting text vectorization method to classify the text and extract the subject.Research one has labeled more than 1200 therapists' responds as high-and low-quality.By adopting those data and vectorize those in different manner,comparing the result of different classifier with different vectorization manner.The result revealed that the effect of different classifier was similar with each other,support vector machine was a little better with respect to F1 measure index.Even though,the result of those classifiers was no worse than peer study,but still have some more distance to put it into practice.This result indicate that computer-aid therapist training is possible,for there is a chance that machine could learn how to give corrective feedback to the trainees who are practicing their respond skills.According to the result of subject extraction,the words that therapists used do reflect client's main theme of counseling sessions to some extent.The result of text clustering could reflect therapists'pet words somehow,but due to the incompatibility between the data set and the method,more research on this was needed.