在人工智慧賦予軟體代理人「思考」的互動能力後,如何賦予軟體代理人「情緒」因素,使之更為可信成為學術研究的重要議題。心理學家認為,情緒因素對於人類生活之每一個細節都有重大的影響,如何將抽象的情緒因素加入軟體代理人,用以模擬人類情緒之變化,或探討情緒因素如何影響人類之思考、學習活動,都是相當有趣的研究課題。 本研究探討多項心理學證據、教育理論等的相關研究,提出一個基本情緒空間模型,主要基於向量空間之數學基礎,並定義如何對從事情緒情緒表徵、情緒加成、及情緒淡化等操作。接著,提出眉口表情空間,以眉毛與嘴型的曲度變化組合,以模擬人類臉部之基本表情。並定義基本情緒空間與眉口表情空間之對應與轉換關係,裨使隱而不顯的情緒狀態,都能夠以表情方式呈現出來。 最後並將基本情緒空間與眉口表情空間,結合基於Tok主要架構之設計,與理性情緒心理學之ABC結構,整合設計出可運作的軟體代理人系統。將此系統運用於線上練習中,根據使用者作練習之狀況,以代理人適時回應情緒狀態給使用者,讓使用者得知是否需加強其在各科目上的學習。
It is a popular research issue to give the ability of initiative emotions to software agents. It is commonly agreed by psychologists that emotions have significant influence on the daily lives of human beings. It is interesting to design software agent with abstract emotion factors, to mimic the emotion variation of human beings, and to know how the emotions take effect to the thinking and learning. The paper proposes a model of basic emotion space on the basis of researches of psychology as well as educational theories. Based on the vector space theories of mathematics, this thesis formulates a set of mathematic operations of emotions, such as the representation of emotions, the combination of emotions, and the fading of emotions. Afterwards, a facial expression space is proposed, which named eyebrow-mouth space, in which various basic facial expression of human can be constructed with different curves of eyebrows and mouth. Since emotion space and facial expression space are set up, a mapping relation of them is defined as well. This helps the implicit emotion states inside the agent become explicit facial expressions. Inspired from Tok software agent architecture and ABC architecture in rational-emotion psychology, the thesis designs an operational implementation system. In this system, an emotional agent works with a user during an examination. According to the performance of this user, the agent gives corresponding facial expressions as feedback to tell the user how much more effort should be paid to the subject.