Previous speech entrainment studies have shown disagreement in their findings: One group emphasized that acoustic entrainment predicts social adaptation, whereas another group emphasized that it predicts social maladaptation. Our study aims to resolve the disagreement from the perspective of emotional entrainment: the entrainment of positive emotions predicts social adaptation, whereas the entrainment of negative emotions predicts social maladaptation. Using a machine-learned sentiment classifier, we estimated the probability of anger, disgust, fear, happiness, neutrality, and sadness in speech. The corpus consisted of dialogues recorded from 29 comprehensive mental health interviews. The Jensen- Shannon divergence was also calculated to estimate the (dis)entrainment. Results showed that the entrainment of happiness significantly demonstrated the rapport of the participants with their therapist. In contrast, their entrainment of disgust significantly demonstrated their social maladaptation. Our study observed social maladaptation to be contrastingly related to positive and negative emotional entrainment. Classification of speech from an emotional perspective could enrich the study of entrainment and facilitate the analysis of emotional communication.