Aspect-Based Sentiment Classification for Detecting the Cognitive Triad Mechanism of Depression

Abstract

The cognitive triad mechanism of Beck’s cognitive theory is critical for early diagnosis and prognosis of depression. According to the cognitive triad mechanism, negative views about the self, the future, and the world appear to be routine in depressed people, occurring spontaneously. The challenge in the clinical interview is that the individuals may have trouble explaining their past and current symptoms for a variety of factors, such as the existence of concurrent mental health or medical problems and memory difficulties. The aspect-based sentiment classification technique can help psychologists overcome this difficulty by identifying the cognitive triad pairs {(self, negative), (future, negative), (world, negative)} from the individual's clinical chatbot and social media messages. The proposed multilayer RNN-capsule architecture on Cognitive Triad Dataset (CTD) consists of two layers, i.e., sentiment recognition and aspect identification layers. Through experiments on the CTD, the multilayer RNN-capsule model outperforms the single-layer RNN-capsule model for aspect-based sentiment classification. The accuracy and F1 score of a single-layer RNN-capsule model for aspect-based sentiment classification are 0.857 and 0.858, respectively. The accuracy and F1 score of the sentiment recognition capsule in the multilayer RNN-capsule model are 0.89 and 0.892, while the accuracy and F1 score of the aspect detection capsule are 0.957 and 0.956. Also, the multilayer RNN-capsule model outperforms most of its counterparts like GNN, LSTM, and BiLSTM. More importantly, capsule architecture is capable of producing words containing sentiment and aspect inclinations that reflect the attributes of sentiment and aspect capsules, respectively, without the use of any linguistic knowledge.