Trouble Sentence and Emotional Sentence Extraction System for Problem Recognition in Psychotherapy Dialogue

Abstract

Recently, as the number of patients with mental illnesses has increased rapidly, more psychotherapists are needed. However, it is still not common to receive appropriate treatment due to high consultation costs and aversion to psychotherapy. This leads to increase of suicides originating in deterioration of mental condition. At the moment, no dialogue system has achieved therapeutic skills close to a human counselor. Therefore, to realize a dialog system capable of recognizing mental problems, we developed three types of methods for extracting trouble sentence and emotional sentence automatically from online Q and A site. Based on these methods, we have developed rule-based extraction system, similarity-based extraction system, and classifier-based extraction system. In the performance evaluation experiments conducted, classifier-based extraction system showed the best F-score of 0.73 when compared to other systems.