「社群情緒指標」於房地產市場價格關聯之研究

Abstract

面對日益高漲的房市價格，民眾對於買房的目標越來越遙不可及，每個人都期望能夠有預知能力能夠提前得知房價漲跌趨勢，以此擬訂購屋策略；相反地，對於賣方市場而言，期望價格能夠越趨成長。目前針對台灣房地產價格趨勢預測，多從總體經濟及各項市場指標來著手，本研究希望藉由網路上的龐大社群網友的情緒擷取，來找出與房地產市場價格的關聯性，期望透過本研究，能夠發展不同的預測方式，提供房市買賣雙方穩定而透明的價格趨勢參考。 本研究選定高人氣的電子網路看板，透過意見探勘的技術，萃取出每位網友針對房地產買賣所分享的資訊與評論中所表達的情緒，並仿照網路評分的方式加以量化，結合成社群情緒指標，分析與房價趨勢之間的關係，並進一步預測內政部營建署「房價綜合趨勢分數」。 研究結果顯示，社群情緒指標與房價趨勢呈現負相關性，當社群情緒為負向情緒時，房價為向上趨勢；若社群情緒為中性情緒時，房價趨勢會反轉向下。在預測內政部營建署「房價綜合趨勢分數」方面，本研究預測模型的預測誤差值均在6%以內，表示本研究之預測模型具有優秀的預測能力。

In the wake of rising house prices, common people's goal to own a house is becoming more distant. For the buyer side, everyone hopes to obtain the ability to predict real estate price trends, and to develop housing strategy based on the prediction. On the contrary, the seller side expects a growing real estate price trend. Currently real estate price trend forecasts in Taiwan are mostly based on the overall economy and wide market indices. This research aims to find relationship of real estate price trend by capturing vast online social network communities' user emotional input, in the hopes to create a different prediction formula and hence to provide a clearer reference for the house market pricing trend. This research selects highly popular electronic discussion board to extract target audience's sentiment regarding real estate transaction information and criticisms, mimics PTT scoring mechanism to quantify it, and combined into “Social Mood Index” to analyze the relationship with the trend of house price and further to predict the “Real Estate Price Trend Score” of Construction and Planning Agency of Ministry of the Interior(CPAMI). This research indicated that there is negative correlation between “Social Mood Index” and the trend of house price. When social mood is negative, the trend of house price appears positive; On the other hand, when social mood is neutral, the trend of house price is negative. In terms of predicting the “Real Estate Price Trend Score” of CPAMI, the deviation value of the prediction is all below 6% on this prediction model, indicating that the prediction model of this study has the good predictive ability.