基於新聞字詞漲跌極性之股價趨勢分類預測

Abstract

根據調查台灣人投資比例排名全球第三，投資工具又以股票投資為主，然而選擇股票標的參考依據有很多種，其中一種方法就是針對財金新聞做分析。隨著網路科技的發達，投資人可以接觸到許多財金新聞，並看到某些新聞上的關鍵字詞，進而直接影響投資人的買賣決策。例如關於台積電新聞出現「出現史上最長排隊潮，訂單塞到爆」這句話，「訂單塞到爆」這個關鍵詞使投資人看好台積電的未來，然後買入台積電股票，影響股價上揚。因此本研究假設財金新聞的關鍵字詞會影響投資人的預期心理與買賣操作，使得個股股價產生漲跌。利用文字處理方法擷取新聞關鍵字詞，判斷每個關鍵字詞對投資人產生正向/持平/負向的影響，然後預測未來股價趨勢。本研究以SVM 演算法實作預測模型，以週為單位結合文字特性之分析與技術指標，探討新聞對於股價漲跌之反應時間與影響，並驗證股價資訊結合文字探勘之模式是否能有效應用於個股漲跌趨勢預測問題，希望提供投資人更有效率的決策資訊。

According to the survey, Taiwanese investment ratio ranked top third in the world, and stocks are regarded as the first investment tool in Taiwan. However, there are many methods to choose what stock investors want to invest. One of methods is the financial news analysis. With the advancement of information technology, investors can catch many financial news articles. When investors read the particular keywords on financial news articles, it would directly affect investors' trading decisions. For example, there is a sentence “There are full of orders, which result in the longest queue in history” about TSMC news. So, “full of orders”, this keywords will make investors think that TSMC have a prosperous future. Then investors would buy the TSMC stock, thus the stock price rises. Therefore, we assumed that keywords of financial news articles would affect investors' psychological expectations and trading operations, then resulting in the stock price up or down. In this study, we take advantage of text processing methods as ways to extract the keywords. Next, we determine the positive, negative or neutral sentiment of each keyword to investors. Finally, we predict future price trends. SVM is the prediction model adopted in this study. We combine the weekly news articles and technical indicators to explore the news reaction time and the influence of stock price up or down. We also evaluate whether combined text analysis and stock price would impact the trend of stock or not. Through our study, we would provide investors an effective investing advice.