- 學位論文
整合TF-IDF與單純貝氏分類器於語意推論基於雲端運算
Integrating of TF-IDF and Naïve Bayes Classifier in Semantic Inference Based on Cloud Computing
摘要
根據「國家發展委員會」的人口推估，在2018年台灣將會進入高齡社會，政府推動長照2.0計畫因應此趨勢，但長期照顧的基礎服務人力不足是目前長照2.0所面對主要問題之一，如何針對長照個案提供正確合適的長照服務以及自動推薦適合的照顧服務員，以善用有限的服務人力資源是本研究所要探討的議題。本研究提出整合統計與機器學習於雲端語意架構(Integrating Statistics and Machine Learning into Cloud Semantic Framework, ISMLCSF)，透過Term Frequency-Inverse Document Frequency(TF-IDF)、Naïve Bayes Classifier和Open Data作為系統基本架構並應用於Semantic Web架設於雲端運算，進行分散式運算以解決龐大資料。本研究以長期照顧為基礎，提出雲端長照平台(Cloud Long-Term Care Platform, CLCP)，以TF-IDF統計方式，應用於照顧服務員的服務項目，計算出該照顧服務員在各項服務的專業度；長照個案的服務需求，依據照顧管理專員評估其他長照個案的歷史量表，透過單純貝氏分類器，自動分類長照個案可能所需的長照服務項目，這些服務項目可轉換為類別物件整合至語意網技術達到推論效果，讓長照個案皆獲得合適的照顧服務，進而讓長期照顧人力資源可以有效的利用，個案家屬也可以依照Open Data所提供的長期照顧機構，了解自己附近的有哪些機構可供使用，並且比較單純貝氏分類器在雲端運算上使用不同的叢集管理器Standalone、Yet Another Resource Negotiator(YARN)、Mesos的執行效能，以及分析其結果所發生的原因。透由CLCP所產生的推薦結果，以此來驗證本研究提出的ISMLCSF可以改善目前長照基礎服務人員不足的問題。
並列摘要
According to the population estimates by the National Development Council, Taiwan is expected to enter the aging society in 2018. To cope with such trend, the government promotes the 10 Years Long-Term Care 2.0 Plan. However, the inadequate manpower of long-term care basic service is one of the main issues confronted by the Plan currently. This study is going to explore how to provide correct and appropriate service for the long-term care cases and recommend the appropriate caring personnel automatically, as well as make good use of limited service manpower. It proposes Integrating Statistics and Machine Learning into Cloud Semantic Framework(ISMLCSF). Through TF-IDF, Naïve Bayes Classifier and Open Data, it is taken as the basic system architecture and applied in Semantic Web set up on cloud computing for distributed computation, so as to resolve the big data issue. Based on the long-term care, this study proposes Cloud Long-Term Care Platform(CLCP). By the means of TF-IDF statistical method, CLCP is applied in the service items provided by the nurse aide, so as to calculate the proficiency of the nurse aide in terms of various service items. As for the service requirements of the long-term care cases, it uses the Naïve Bayes Classifier to classify the long-term care service items possibly needed by the cases automatically based on the history scale evaluated by the care manager for other long-term care cases. These service items can be converted into category subjects and integrated into semantic technology to achieve inference effect. In this way, all long-term care cases can receive the desired service items, furthermore, the long-term care manpower resources can be utilized efficiently. The guardians of the cases can learn the available institutions nearby based on the list of long-term care institutions provided by Open Data. With the recommendation results generated by CLCP, it can validate the ISMLCSF proposed by this study can improve the issue of short hands for long-term care basic service.