Predicting Taiwan Stock Market Using Social Moods

近年來有許研究再探討透過分析社群網站上的資訊來預測未來的可能性當然，不管在什麼時期，投資者的情緒和股市的波動常常被放再一起討論，而隨著社群網站的快速發展，越來越多人願意在網路上宣洩他們的情緒，當然投資者也不例外，因此藉由分析他們的情緒來預測股市成為近幾年的熱門議題。 在這篇論文研究中，PTT是我們要分析的社群網站，有許多投資者在PTT的股市版上發表的他們對於股市的看法還有給予其他投資者的建議，透過分析裡面的資料可以了解到台灣投資者的行為與情緒，我們透過台灣大學編訂的NTUSD跟大連理工大學編定的DUTIRSD這兩個情緒字典來量化投資者的情緒並預測台灣兩個代表性的股市指數，分別為台股期貨指數(TX)與台股加權指數(TAIEX)。我們利用的移動視窗的概念並在每一個移動式窗內選取固定數量的特徵來進行預測，我們認為隨著時間的變化，造成股市震盪的因素也會改變，因此不應該選特定特徵來預測未來的每一天，我們透過均方根誤差(RMSE)來決定移動視窗的大小和選定特徵的數量，越小的RMSE代表當下的移動視窗大小和選定特徵數量有較大的預測能力。 每一天的股市都記錄著開盤價、最高價、最低價、收盤價，我們透過K-means分群演算法將四個價格區分成三類並透過投資者的情緒來預測隔日的股市的四個數值分別屬於哪一種狀態。In recent years, mining social media data to forecast the future has been a popular research. The stock market behavior and investor emotions are always bonded together. With the development of social media, people are willing the share their feelings on the social media including investor. In our study, we select PTT stock board as our platform, a forum gathering investors sharing their opinions, and crawl data on it. We calculate the emotion score through NTUSD and DUTIR sentiment dictionary and predict two representative stock market indices: Taiwan Futures Index and Taiwan Capitalization Weighted Stock Index. The concept of fixed-sized rolling window and fixed feature size are adopted in this thesis. That is, if the emotion cause the variation of stock market, the main causality might be different in different time span. The rolling window size and feature size are selected to our prediction model through lower Root Mean Square Error. There are four value recorded each day: opening value, intra-day highest value, intra-day lowest value and closing value. We classify these four value into three groups through K-means clustering algorithm and then conduct prediction