使用深度學習技術之具有同理心的情感支持系統

Abstract

情感支持是一項極為重要的技能，它能安慰正處於低潮的人。在現代高壓的社會之中，人們經常被各種各樣的壓力逼迫得喘不過氣。通常，我們能夠從朋友或家人的身上尋求情感支持，藉此度過自己艱難的時期。但實際上，不可能每次我們需要時，都有人在身邊提供幫助。在本篇論文中，我們提出了一個能夠根據使用者所說的話來產生情感支持回覆的系統。此系統使用深度學習的技術，以從使用者的話語中提取情緒和語意資訊。在接收到使用者的話語之後，系統會先辨識語句的情緒。根據不同的情緒，系統會調整使用不同的方法來產生適當的回覆給使用者。 同理心是指能夠理解別人的經歷、分享別人感受的一種能力。若要提供有效的情感支持，同理心是不可或缺的。若要表現出同理心，我們必須了解目標對象的情緒以及其產生該情緒的原因。除此之外，若能對其經歷的情況具備一定的知識，也能夠使我們更加了解對象所經歷的事情、進而提供更好的情感支持。在我們提出的系統中，設計了一個能夠根據使用者輸入，整合使用者情緒、情緒產生原因、額外知識等資訊，並產生具有同理心回覆的深度學習模型。

Emotional support is an important skill that comforts people when depressed. In the stressful society, people are often overwhelmed by various pressure. Usually, emotional support from our friends or family can help us get through the tough time. However, it is unlikely to have people constantly provide emotional support when we are in need. In this thesis, we propose a system that can automatically generate emotional support responses according to the utterance from the user. The system uses deep learning techniques to extract both emotional and semantic information of the user’s utterances. After receiving the utterance from the user, the system will first recognize the emotion of it. According to different emotion, the system will adapt its strategy to generate an appropriate response to the user. Empathy is the ability to understand other people’s experience and share their feelings, it is necessary if we want to provide effective emotional support. To show empathy, we need to understand the recipient’s emotion meanwhile the cause of the emotion. Moreover, having the background knowledge about the recipient helps us understand the recipient’s experience and provide better emotional support. In our proposed system, we design a deep learning model that integrates the information of user’s emotion, its cause and external knowledge to generate empathetic responses to user’s utterance.