Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/73128
Title: | 檢測房價預期對房屋市場之影響 - 利用文字探勘技巧分析網路論壇民意 Do House Price Expectations Affect the Housing Market? Using Text Mining Techniques to Analyze Public Sentiments on a Social Media Forum |
Authors: | Hua-Hsing Huang 黃華興 |
Advisor: | 陳南光(Nan-Kuang Chen) |
Co-Advisor: | 王泓仁,盧信銘(Hsin-Min Lu) |
Keyword: | 房屋市場,文字探勘,情緒分析,機器學習,因果檢定, Housing Market,Text Mining,Sentiment Analysis,Machine Learning,Granger Causality, |
Publication Year : | 2019 |
Degree: | 碩士 |
Abstract: | 本篇論文嘗試去調查台灣民眾情緒與房屋市場的因果關係。相較於傳統文獻主要利用問卷調查方式蒐集民眾的意見,本篇論文以網路論壇的資料為基礎,透過機器學習與情緒分析的方法建構出新的情緒指標。我們從 mobile01 論壇中擷取了超過 160 萬篇的貼文,並考慮三種不同的機器學習模型來建立模型。結果顯示三元分類的支援向量機模型(SVM)在訓練資料集中表現最佳,因此本篇論文即採用該模型作為預測模型。最後,透過 Granger 因果檢定,我們發現了三個不同的因果關係。第一,交易量 Granger-causes 房屋價格;第二,成交量 Granger-causes 民眾情緒;最後,民眾情緒 Granger-cause 房屋價格。上述三種單向的因果關係可能提供了交易量如何影響房屋價格的部分解釋。 The paper investigates the causal relations between the public sentiments and the housing market in Taiwan. Instead of using survey data to measure the sentiments like most of the studies, the paper constructs the sentiment indices by narrative analytical methods. We scrape more than 1.6 million posts from Mobile01, which is one of the biggest social media forums in Taiwan. As the ternary support vector machine (SVM) model outperforms other machine learning models in our data, it is used for predicting the sentiments of the posts. Through Granger causality test, we find three different causal relations. First, quantities Granger-cause house prices. Second, quantities Granger-cause the sentiments. Finally, the sentiments Granger-cause house prices. The results of these unilateral Granger causalities provide a hint for the mechanism of how transaction volumes affect house prices. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/73128 |
DOI: | 10.6342/NTU201901339 |
Fulltext Rights: | 有償授權 |
Appears in Collections: | 經濟學系 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-108-1.pdf Restricted Access | 2.35 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.