就股票投資決策而言,投資者會從外在的眾多媒體中,找尋可能的投資資訊。先前研究大多採用技術分析法或基本分析法,或是運用類神經網路做股價趨勢的預測。但基本分析法對短期的市場變動較不敏感,且精確度較低,技術分析只關心證券市場本身的變化,而不考慮會對其產生某種影響的經濟方面、政治方面的等各種外部的因素,而先前運用類神經網路的研究中,多只採用數值型態的指標或歷史股價作為分析來源。資訊時代中,越來越多的新聞以電子化型式呈現,對投資者而言,新聞資料除了包含著隱藏的投資資訊,這些資訊也會反應到股票市場中,因此如何能有效的分析這些非結構化的資料,是目前亟需研究的議題之一。 以資料探勘技術來預測股價趨勢在以往已有大量的的研究,但大多是使用數值資料作為分析的來源;此外,雖有少量的研究以非結構化資料(新聞文件、專家評論、網路論壇等)來預測個股當日的股價趨勢,但不論是精確度或是準確度上仍有待突破,且多是國外股市之預測模型,並不一定適用於台灣股市環境。 本研究整合股價資料與中文財經新聞,建構一預測台灣股市日內股價漲跌趨勢之模型。以往類似研究通常只採用歷史股價做基本分析或是技術分析來預測股價之漲跌趨勢,或是大多採用英語系之新聞作為模型之輸入;本研究與先前研究之差異點為透過詞性標記與詞性組合規則策略萃取關鍵字,研究除了驗證該模型於中文文件預測台股個股漲跌趨勢可行外,並以提高整體預測正確率為目標。
Investors will look for probable investment information from numerous electronic publishing that to make investment decision. More and more news presented by electronic format in the information age. The immediate news includes investment information besides will response to stock market. How to deal with the mass unstructured data efficiently is an important issue. Stock price trend forecasting based on structured data enjoys great popularity, but no matter using the statistics methods or machine learning algorithms still use structured numerical data. Though have a good few study in using unstructured data to predict the future of stock prices trend, the accuracy yet low and the model apply to Taiwan Market is uncertain. In this paper we integrate stock tick data and Chinese finance and economics news, to construct a model that can predict intraday stock price trends in Taiwan Market. In general, prefer using Fundamental Analysis or Technical Analysis to forecast stock, and furthermore mostly use English news to regard as input of prediction model. The main difference between the past researches is that combine the part of speech tagging and heuristic rules to extract keywords. In addition to prove the prediction model is suite for Taiwan Market, also increase the accuracy of prediction is the chief objective. The framework include two phase, first is building and training model, then via experiment to probe into the variances with the effect of the accuracy, to induce the best settings of prediction model. Finally, through the simulation trading experiment to evaluate the feasibility and practical of model.