電機資訊學院: 資訊工程學研究所指導教授: 傅立成蔣宗圻Chiang, Tsung-ChiTsung-ChiChiang2017-03-032018-07-052017-03-032018-07-052015http://ntur.lib.ntu.edu.tw//handle/246246/275440隨著物聯網的科技趨漸普及與流行，使得低運算能力的裝置在嵌入相應的感應器與控制器後，大量的佈建於智慧環境中。此外，這些感應器所產生的大數據可被進一步的分析，並應用於以人為本更人性化的系統中。其中一個具代表性的應用系統就是智慧家庭，這類應用利用機器學習或資料探勘的方式分析所收集的大數據，根據分析的結果建立情境感知系統於辨識使用者的活動與相對應的服務設定。但是，智慧家庭中的大數據隨時隨地的產生，資料量不斷得成長，使得傳統的監督式學習必須使用大量的人力標註每個資料的意義而不易應用於現實環境中。此外分析結果所建立的模型必須精簡且精確，而能在低運算力的裝置中運行並且佈建於智慧家庭。因此本研究提出一個非監督式學習分析方法於多人的環境中的情境探勘與相對應的服務參數，導入適合智慧家庭特性之情境結構以降低演算法複雜度同時不失正確率，並提供多人情境下滿足使用者偏好最佳化之服務。 本研究的主要貢獻有以下三點: 第一，為了有效的分析智慧家庭所產生的大數據與降低人為的資料標示成本，我們提出一個非參數式的模型用以探索家中使用者之行為。第二，我們提出一個非監督式的方法學習使用者於智慧家庭中的服務偏好參數。第三，為了主動地提供最適合的服務，我們提出一個服務決策系統，並根據探索而得的使用者當下行為與對應的服務參數提高系統提供服務時使用者的滿意度。As we are moving towards the Internet of Things (IoT), a significant number of heterogeneous sensors deployed around the word. The amounts of Big Data are contin-uously generated from those sensors in activity daily living (ADL). One of the applica-tions in IoT is smart home that proactively provides context-aware services by recog-nizing user activities among sensory data. In the other word, activity recognition (AR) plays a key component to add value to raw sensor data we need to understand it. Collec-tion, modeling, reasoning, and distribution of context in relation to sensor data plays critical role in this challenge. Context-aware computing has proven to be successful in understanding sensor data. However, for AR to build personalization activity daily liv-ing for each user and overcome the human labeling effort in Big Data, it is important to exploit unsupervised way from activity discovery (AD) [2], which is finding unknown patterns directly from low-level sensor data without any pre-defined assumptions. There are three major contributions in this thesis. Firstly, a novel nonparametric model of activity discovery is proposed to reduce the effort of data annotation while the analysis of Big Data is being performed. Secondly, we propose a service discovery en-gine to learning user preference in an unsupervised way. Thirdly, a decision engine for proactively providing suitable service is proposed. We combine both the discovered user activity and corresponding service preference to improve the suitability of service with smart environment.1924872 bytesapplication/pdf論文公開時間: 2020/8/21論文使用權限: 同意有償授權(權利金給回饋學校)智慧家庭行為辨識行為探索服務探索非監督式學習Smart HomeActivity RecognitionActivity DiscoveryUnsupervised Learningthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/275440/1/ntu-104-R02922097-1.pdf