Title page for etd-0202121-134610
 論文名稱Title 基於多模態學習的房產鑑價模型 ── 以高雄市為例House Price Prediction based on Multimodal Learning －A Case Study of Kaohsiung City 系所名稱Department 資訊管理學系Department of Information Management 畢業學年期Year, semester 109 學年度 第 2 學期The spring semester of Academic Year 109 語文別Language 英文English 學位類別Degree 碩士Master 頁數Number of pages 41 研究生Author 閻俞蓉Yu-Jung Yen 指導教授Advisor 召集委員Convenor 口試委員Advisory Committee 口試日期Date of Exam 2021-02-22 繳交日期Date of Submission 2021-03-02 關鍵字Keywords 坡度提升演算法、綠地、多模態學習、Google 街景、房屋價值、隨機森 林演算法Random Forest, XGBoost, Green space, Google Street View, House Price, Multimodal learning 統計Statistics 本論文已被瀏覽 454 次，被下載 112 次The thesis/dissertation has been browsed 454 times, has been downloaded 112 times.
 中文摘要 人們可以輕易辨識圖片的內容，從房屋外觀分析出該房屋的價錢、屋齡等， 機器也可以在辨識度上有很好的結果。本研究以多模態學習為基礎，取用內政部 不動產成交案件實際資訊資料供應系統提供的資料，加上利用 Google 街景服務取 得每個地址的街景圖。使用 XGBoost 演算法對圖片做處理，並利用隨機森林 (Random Forest)、線性迴歸等演算法，建立模型，預測房價。依據全台六都的都市發展計劃，高雄市公告公園綠地佔都市計畫區面積百分 比居於首位，因此本研究除了探討房屋本身的條件如何影響預測房價的結果，同 時使用 Google distance API，探討公園綠地此一變數對房價呈現多大的影響。實證 結果，圖片本身能夠達成預測房價，加入房屋原始資料後，提升預測準確率。此 外，實驗證明，700 公尺內擁有越多公園綠地的房屋，顯著呈現高房價。 Abstract People can easily view the picture to judge the price of a house through the building age, environment conditions and appearances. Machines can also reach a good result as humans do. This study focuses on multi-modal learning by using the actual transaction information from the Ministry of the Interior’s real estate data system and the street view images downloaded from Google Street View service. We convert images by XGBoost and add new converted data to build a new model to predict house prices.According to Taiwan government’s urban development plan, Kaohsiung City possesses the largest percentage of green space coverage compared to the other five special municipalities of Taiwan. Therefore, to discover how green space affects the prediction of house prices, this study also used Google distance to discuss the importance of green space. From the empirical results, the accuracy of combining the transaction data with images is improved. Furthermore, our experiments show that the distance within 700 meters from the house to green space, the influence of the green space on house prices is significant.
 目次 Table of Contents 論文審定書 i 摘要 ii Abstract iii List of Figures v List of Tables vi 1. Introduction 1 2. Background and Related Works 22.1. Related Work on House Prices 2 2.2. Multimodal Learning 4 2.3. Google Street View 6 2.4. Ensemble learning 8 2.5. EXtreme Gradient Boosting (XGBoost) 9 2.6. Random Forest 103. Methodology 11 4. Experiment and Discussion 14 4.1. Environment setup 15 4.2. Transaction Dataset 15 4.3. Green Space Feature 16 4.4. Images Feature 17 4.5. Performance Comparison 18 4.6. Important Variables 21 4.7. Discussion 24 5. Conclusion 24 Reference 26 中文文獻 33 Appendix A 34