具個性提示之聊天機器人對電子商務使用者感受與行為之影響

隨著人工智慧逐漸成熟，聊天機器人(Chatbot)也隨之崛起，能提供多樣化的服務，在生活中逐漸扮演重要的角色。過去研究指出，在電腦人機互動時，人們容易將電腦擬人化並賦予個性，且擁有社交與信任關係，而人們與具個性的Chatbot互動時，感受及行為也產生了差異，因此Chatbot的個性感知與使用者信任感受之間的關係，也成了值得進一步探討的議題，然而該如何塑造Chatbot個性，來增強使用者的信任感受進而有效地完成任務呢，在人際互動時，人們會從外觀來判斷他人個性，同樣在電腦人機領域中，學者認為視覺外觀也是個性感知的其中一條線索。本研究希望了解Chatbot在與人們互動時，具外觀(或稱頭像)個性提示是否影響參與者對於Chatbot的感受及行為。本研究分為兩部分，第一部分是前導研究，透過案例分析歸納出目前Chatbot頭像常用的設計元素：髮型、髮色、眼睛細緻度、抽象程度、圖像底色及眼鏡有無，以田口式直交表實驗法選出八個不同的元素組合作為代表樣本，透過五大個性特質的網路問卷，進行樣本個性感受調查，藉由數量化一類來分析回收之202份問卷，了解不同設計元素對Chatbot個性感知的影響，並找出計算Chatbot個性分數的複迴歸公式。第二部分為正式實驗，透過前導研究所選出最具備「盡責」、「外向」、「不盡責」與「內向」四種個性的Chatbot頭像作為自變項，在Facebook Messenger訊息平台上以商品推薦作為任務情境來和參與者進行線上互動，一共招募240位參與者且平均分配於四種實驗條件中，採組間設計，以問卷與點擊率作為測量參與者信任感受與行為的測量方法。研究結果發現：(1)參與者會因頭像設計元素的不同而感知Chatbot個性的差異；(2)參與者會受Chatbot個性不同的影響而產生不同的感受及行為；(3)外向個性的Chatbot使參與者產生較高的信任感受及點擊率；(4)內向個性的Chatbot使參與者產生較低的信任感受及點擊率。未來在設計Chatbot時若將頭像個性提示納入考量，將是其中一個提升信任的途徑，最後依據研究過程提出幾點未來研究建議。

As artificial intelligence matures, chatbots also rise. Chatbots offer a variety of services and gradually play an important role in people's lives. Studies have shown that in human-computer interaction, people tend to anthropomorphize computers and give them personality, and even develop social relationships or build trust. When people interact with chatbots, their personalities could have an impact on people’s perceptions and behaviors. How to shape a chatbot’s personality to enhance the user's trust and task efficiency has become a design issue worthy of further investigation. Therefore, this study aims to investigate the effect of chatbot’s personality cues on users’ perception and behavior in an electronic commerce (e-commerce) context. The study is divided into two parts: pilot study and experiment. In the pilot study, six design elements commonly used in the current chatbot’s avatars were generalized, namely hairstyle, hair color, degree of eyes detail, abstraction level, background color, and with/ without glasses. Then, eight chatbot avatar samples combining the six above-mentioned elements were selected using the Taguchi’s orthogonal arrays. Via the online questionnaires based on the Five-Factor Model (FFM) of personality to collect participants' perception of each sample. The 202 questionnaires collected were analyzed by Quantification Theory Type I to understand the influence of different design elements on chatbot’s personality and to build the multiple regression equation that estimated the chatbot’s personality score. The second part is the formal experiment. Through the pilot study, the four chatbot avatars with the highest estimated score in personalities of "conscientiousness", "extraversion", "low conscientiousness" and "introverted" were selected as independent variables. Each participant interacted with one product recommender chatbot on Facebook's Messenger platform and received online product recommendations from it. A total of 240 participants was recruited and distributed equally among the four experimental conditions. The participants' perception of trust was measured with a questionnaire, and the click-through rate was used as a measure of participants' behavior. According to the study results, the following conclusions are drawn: (1) Participants perceive the personality differences of chatbots based on different avatar design elements; (2) Participants would have different perceptions and behaviors when perceiving different personality of chatbots; (3) Participants have a higher perception of trust and click rate for the chatbot avatar with extraversion personality; (4) Participants have a lower perception of trust and click rate for the chatbot avatar with introverted personality. Finally, implications for design and suggestions for future work in this area are provided based on this study.