基於時空和文本資料之感測器相關對話機器人

Abstract

近年來，隨著自然語言分析(NLP)和人工智慧(AI)的成長，對話機器人(chatbot)變得越來越流行，漸漸成為人們用來獲取資訊的良好介面。有許多針對不同目的的對話機器人被實作出來，但很少對話機器人可以連結感測器、動態的查詢其相關資訊。有鑑於此，我們在Facebook messenger上開發了一個可以查詢感測器動態資訊的對話機器人，並從不同的角度對感測器相關問答進行了分析。我們分析感測器相關問答並提出了「黃金三角」，分別是感測器類型、空間因子和時間因子，這些是查詢感測器數據的三個基本要素，有了這三大要素我們才可以取得想要的感測器資料。使用者可以用自然語言直接詢問感測器的動態資訊，我們將從中分析這三個要素，並動態的取得感測器的數值來判斷目前狀態並回傳答案給予使用者適當的建議。我們的對話機器人更支持許多有趣的問答方式，突破傳統使用者只能一步步隨著規則走的標準問答框架。此外我們還提供了訂閱功能，使用者可以輕鬆追蹤他們在意的感測器數值，當數值超過標準時也會立即給予警報。此研究中展示了我們的對話機器人實際上如何與使用者互動、回答使用者對於感測器資訊的相關問題，並進一步分析目前取得的成效和使用者的回饋。

Chatbots are becoming increasingly popular along with the growth of NLP and AI these days. They are developed for different purposes, but seldom are they related with sensors. In this work, we built a sensor-related chatbot on Facebook Messenger and analyzed sensor-related question answering from different perspectives, in order to gain a holistic view of it. We propose the concept of a golden triangle of sensor-related question answering, consisting of sensor type, spatial factor and temporal factor, three essential factors for querying sensor data. Users can ask sensor-related questions in natural language, and we will extract these three factors and give the answer based on dynamic sensor values. Our chatbot also provides a way to support many more interesting questions users may ask, and can break through the traditional standard question answering framework, in which users always need to follow the chatbot step by step. Furthermore, we provide a subscription function, so users can keep an eye on whatever sensor information they care about, and also get instant warnings when the sensor value exceeds the standard.