我們希望透過英語閱讀的方式使學生對英文產生興趣，讓學生透過閱讀將英文與自我生活的產生連結，在社會脈絡中發展語言的能力。然而這樣的社會文化建構過程需要大量的師資，在目前有限的人力資源下並不可行。 因此我們對話系統以聊故事為主軸與使用者建立共同話題並展開對話，我們團隊希望學生與聊天機器人的互動，不只是單純得進行故事的討論，也可以進行日常對話的問答或是讓學生適當地擁有主導的話語權，然而這目前仍然是一個挑戰，因為在多面向模組的整合下，機器人更需要有充足的自然語言理解以及對話策略選擇的能力，可以自動且有效率的提供符合當下情境的回應。 此文的主要任務就是要介紹我們如何訓練一個教育對話機器人模型，讓他可以從多種狀態下去察覺學生的情況，再探勘此狀態組合的對應的回覆，在此模型中我們採用了強化式學習(Reinforcement learning)的訓練架構進行訓練，以此達到此論文最終目的---與使用者建立關係並使對話長久進行。;We hope that through reading in English, students will be interested in English, so that students can connect English with their own life through reading, and develop their language ability in the social context. However, such a social and cultural construction process requires a large number of teachers, which is not feasible under the current limited human resources. Therefore, our dialogue system takes the story as the main axis to establish a common topic and start a dialogue with users. Our team hopes that the interaction between students and chatbots is not only a simple discussion of stories, but also a question-and-answer session in daily conversations or allowing students to appropriately However, this is still a challenge, because under the integration of multi-faceted modules, robots need to have sufficient natural language understanding and dialogue strategy selection capabilities, which can automatically and efficiently provide products that meet the needs of the current situation. situational response. The main task of this article is to introduce how we train an educational dialogue robot model, so that it can detect the situation of students from various states, and then explore the corresponding replies of this combination of states. In this model, we use Reinforcement learning training architecture to achieve the ultimate goal of this paper - to establish a relationship with the user and make the dialogue perpetual.
