深度學習與情感分析應用於股價預測

Stock Prediction Using Deep Learning and Sentiment Analysis

盧奕叡(Yi-Jui Lu)

指導教授 ： 孫春在

國立交通大學/資訊學院/資訊科學與工程研究所/碩士(2018年)

摘要

多年來股價預測不管在學術界或是業界都是十分重要的議題，過去已有許多研究利用歷史交易資料以及各種技術指標搭配深度學習的技術來提高股價預測的準確率，而近年來有一些研究發現，利用自然語言處理的方法來分析網路上大量的文字資料(財經新聞、網路社群等等)後作為輸入特徵加以預測能夠讓預測的準確率更加提升。
在台灣最大的網路論壇PTT上的Stock版，討論版上除了相關的新聞和股市資訊之外，也會有網友在板上尋求意見或參考他人看法。其中還有特殊的「標的」類文章，網友會針對個股提出自己的分析讓其他網友參考，當中不乏有勝率不錯的網友，其言論甚至比電視上的投顧老師或是證券分析師更加具有影響力，在這些網友發表文章後，對於成交量較低或是價格較低的個股(小型股)其股價可能會被影響，這個現象不禁令人好奇，若是利用這些文章來預測股市是否會有更好的成效。
結果發現，相較於只使用歷史交易資料來預測，加入文章後確實可以使得預測的結果更好，但是股本較小的個股相較於股本較大的個股，其預測結果的提升並沒有顯著的差異。

關鍵字

深度學習 ； 情感分析 ； 股價預測 ； 批踢踢

並列摘要

Stock price prediction has always been an important topic, in the past decades, many researchers used trading history (open price, close price and volume etc.) with deep learning technique attempt to improve the result of the prediction. However, recent years a lot of researchers have found that the text on the social media like twitter, facebook or online news could be used to make the result of prediction more precise and better.PTT stock board is one of the most popular forum in Taiwan, users not only view the information and news of stock market but also seek advices and take opinions of other users. The most interesting part is that there is a special article category called “stock target”, many users post “stock target” article to share their perspective on the specific stock and their analysis. The opinions of some users with high winning percentage are even more influential than stock analysts, so the stock with cheaper price and lower volume (less share capital) may be affected by the article of those users, therefore, in this work we attempt to find out that whether the articles could be used to train a more accurate model and obtain better result of prediction and whether the stock with cheaper price and lower volume is more affected by the articles.The result shows that the model trained by features of sentiment analysis for the articles is more accurate than the model only trained by trading history. The sentiment score of articles could help us obtain a better result of prediction. However, the result also shows that the prediction of stock with cheaper price and lower volume did not improve much more than other stock. There is no significant difference among the stocks with different amount of share capital.

並列關鍵字

Deep Learning ； Sentiment Analysis ； Stock Price Prediction ； PTT

參考文獻

［1］ Längkvist, M., Karlsson, L., & Loutfi, A. (2014). A review of unsupervised feature learning and deep learning for time-series modeling. Pattern Recognition Letters, 42, 11-24.

［2］ Bollen, J., Mao, H., & Zeng, X. (2011). Twitter mood predicts the stock market. Journal of computational science, 2(1), 1-8.

［3］ Sul, H. K., Dennis, A. R., & Yuan, L. I. (2014, January). Trading on Twitter: The financial information content of emotion in social media. In System Sciences (HICSS), 2014 47th Hawaii International Conference on (pp. 806-815). IEEE.

［4］ Shiller, R. J. (1999). Human behavior and the efficiency of the financial system. Handbook of macroeconomics, 1, 1305-1340.

［5］ Fama, E. F. (1965). The behavior of stock-market prices. The journal of Business, 38(1), 34-105.