互動式陪伴型機器人之情感支持系統

Abstract

情感支持是人與人之間協調關係最好的方式，人們在心情低落時，往往需要朋友、家人提供適當的關懷，來幫助自身度過負面的情感。在本論文中，我們開發一個能夠提供情感支持的人機互動系統。該系統能夠藉由觀察人的臉部表情與言語，在使用者是負面情緒時，主動給予關心，並基於使用者的自我披露，提供適合的情感支持回覆。 為了瞭解人的情感，我們應用深度學習的技術結合人臉語言語資訊識別出使用者的情緒，並以此作為主動關懷的觸發點。接著使用自然語言處理的技術來分析使用者的自我披露，並以分析出的資訊配合結合貝氏網路 (Bayesian Network) 的常識知識圖譜來推斷出人的心理情感以及其心情低落的壓力來源。而後應用神經網路 (Neural Network) 從壓力來源的種類推論出合適的情感支持類型。最後基於情感支持的類型和推論出的人之情感產生一個合適且具同理心的對話來與人互動，以此達到情感支持的效果。

Emotional support is a best way to coordinating relationship among people. When people are down in the dumps, appropriate care provided by friends or family usually can help people having upset. In this thesis, we aim to design a robot companion system which is able to provide emotional support to humans. While a human is in a bad mood, the system can generate an appropriate emotional support response based on his/her self-disclosure induced by the robot. To understand human’s emotion, we apply deep learning techniques to recognize the facial expression and analyze the sentence sentiment, which in turn can be used as a cue for robot to proactively care humans. Given the information from the natural language processing as observations, the system can infer human’s feelings and stressors by applying a Bayesian Network which is constructed from commonsense knowledge graph. Afterwards, the system infers emotional support of a specific category from the human’s stressor by neural network. Based on the content of emotional support and inferred human’s feelings, the robot generates an appropriate dialogue which contain empathy to achieve the effective emotional support.