股票是現今社會中多數人喜歡的投資理財工具，如何選擇適當的買賣時機一直都是投資者關心的議題。目前對於選擇投資標的及買賣操作，多數會採用技術面與基本面分析。但消息面對於股市的影響也非常的大，由於頻繁發佈之股市新聞蘊含大量的市場消息，其內容能夠反映不同層面事件對於股市的影響，可能藉此改變投資人預期心理及買賣策略。因此建構一套能分析個股相關新聞以結合技術面分析之模型，透過深度學習技術判斷買賣時機，提供投資者較為準確的買賣資訊，是本論文研究的主題。本研究針對多家上市公司財務資訊以及新聞資訊進行實驗。主要透過了文字探勘演算法篩選出新聞關鍵詞彙及產生對應重要度分數，進而結合詞向量模型訓練之輸出，針對個股計算每日該個股的新聞向量，並將其結合財務資料以長短期記憶(LSTM)遞歸神經網路建立預測模型，預測股價之漲跌幅。最終比較本研究考慮新聞因素之模型(Text-LSTM)、長短期記憶神經網路(LSTM)以及差分整合移動平均自迴歸模型(ARIMA)的預測準確率。結果顯示Text-LSTM在變動方向性(Directional Symmetry, DS)於個股預測平均結果為61.8%，最高為69.3%，正確率分別超出LSTM及ARIMA模型約6.2%及13.4%。

Nowadays, the stock is still the most popular investment tool in our society. To find the time of trading for a stock is the most important issue that concerns all the investors. When selecting an investment target and finding the time of trading, most of investors take advantage of technical analysis and fundamental analysis. However, the influence of financial news on the stock market should also be taken into account since the frequently released financial news carry large amount of information that influences the expectation of an investor on a stock. As such, the financial news may affect the trading behavior of the investor. This thesis aims at constructing a prediction model on the variations of a stock price based on the company-related news and technical analysis of the stock. To this end, we collected financial news of several listed stocks in Taiwan Stock Exchange (TWSE). We first use text mining algorithm to filter out important keywords from the news and calculate their corresponding scores of importance. Then, we train to find the representing vectors of the important words using Word2Vec model. Finally, the vectors of important words of the daily news of a stock are multiplied by their corresponding scores of importance to generate a news vector for the stock. With the daily news vector and several financial variables, we construct a prediction model on the selected stocks using the long short-term memory (LSTM) recurrent neural network. We conducted several experiments on the proposed model, termed Text-LSTM, a model without considering the daily news, termed LSTM, and a model constructed using ARIMA model. The experimental results showed that the Text-LSTM achieved an average of 61.8% accuracy and a maximum of 69.3% accuracy in Directional Symmetry (DS) on the predictions. The accuracy of the Text-LSTM outperforms those of the pure LSTM and the ARIMA by 6.2% and 13.4%, respectively.

[1] Fama, E. F. (1995). Random Walks in Stock Market Prices. Financial Analysts Journal, 51(1), 75-80.
[2] Keynes, J. M. (1936). The General Theory of Employment, Interest, and Money. New York: Harcourt.
[3] Mindell, J. (1948). The Stock Market: Basic Guide for Investors. New York: Forbes.
[4] Graham, B. & Dodd, D. L. (1951). Security analysis: Principles and technique. New York: McGraw-Hill.
[5] Olson, O. & Mossman, C. (2003). Neural Network Forecasts of Canadian Stock Returns using Accounting Ratios. International Journal of Forecasting, 19(3), 453-465.
[6] Thawornwong, S. & Enke, D. (2004). The adaptive selection of financial and economic variables for use with artificial neural networks. Neurocomputing, 56, 205-232.
[7] Chen, Y. J. & Chen, Y. M. (2013). A fundamental analysis-based method for stock market forecasting. Proceedings of 2013 Fourth International Conference on Intelligent Control and Information Processing (ICICIP), 354–359. doi:10.1109/ ICICIP.2013.6568097
[8] Tsai, C. F. & Hsiao, Y. C. (2010). Combining multiple feature selection methods for stock prediction: Union, intersection, and multi-intersection approaches. Decision Support Systems, 50(1), 258-269.
[9] Edirisinghe, N.C.P. & Zhang, X. (2007). Generalized DEA model of fundamental analysis and its application to portfolio optimization. Journal of Banking & Finance, 31(11), 3311-3335.
[10] Lam, M. (2004). Neural network techniques for financial performance prediction: integrating fundamental and technical analysis. Decision Support System, 37(4), 567–581.
[11] Kirkpatrick II, C. D. & Dahlquist, J. R. (2006). Technical Analysis: The Complete Resource for Financial Market Technicians. FT Press. Retrieved from http://1.droppdf.com/files/wWSus/technical-analysis-the-complete-resource-for-financial-market-technicians-2011.pdf
[12] Murphy, J. J. (1999). Technical Analysis of the Financial Markets: A Comprehensive Guide to Trading Methods and Applications. New York: New York Institute of Finance.
[13] 林成蔭(2011)。股海樂活。法意出版社。
[14] 杜金龍(2002)。技術指標在台灣股市應用的訣竅。財訊出版社。
[15] Oriani, F. B. & Coelho, G. P. (2016). Evaluating the impact of technical indicators on stock forecasting. Proceedings of 2016 IEEE Symposium Series on Computational Intelligence (SSCI), 1-8. doi:10.1109/SSCI.2016.7850017
[16] Gholamiangonabadi, D., Mohseni Taheri, S. D., Mohammadi, A. & Menhaj, M. B. (2014). Investigating the performance of technical indicators in electrical industry in Tehran's Stock Exchange using hybrid methods of SRA, PCA and Neural Networks. Proceedings of 2014 5th Conference on Thermal Power Plants (CTPP), 75-82. doi:10.1109/CTPP.2014.7040698
[17] Ni, L.P., Ni, Z.W. & Gao, Y. Z. (2011). Stock trend prediction based on fractal feature selection and support vector machine. Expert Systems with Applications, 38(5), 5569-5576.
[18] Chen, J. F., Chen, W. L., Huang, C. P., Huang, S. H. & Chen, A. P. (2016). Financial Time-Series Data Analysis Using Deep Convolutional Neural Networks. Proceedings of 2016 7th International Conference on Cloud Computing and Big Data (CCBD), 87-92. doi:10.1109/CCBD.2016.027
[19] Bao, W., Yue, J. & Rao, Y. (2017). A deep learning framework for financial time series using stacked autoencoders and long-short term memory. PLOS ONE, 12(7).
[20] Huang, C. L. & Tsai, C. Y. (2009). A hybrid SOFM-SVR with a filter-based feature selection for stock market forecasting. Expert Systems with Applications, 36(2), 1529-1539.
[21] Sullivan, D. P. (2001). Document Warehousing and Text Mining. New York: Wiley.
[22] Salton, G. & McGill, M. J. (1983). Introduction to modern information retrieval. New York: McGraw-HIII.
[23] Sanchez, N. (2012). Communication Process. Retrieved from
https://web.njit.edu/~lipuma/352comproc/comproc.htm
[24] Harris, Z. S. (1954). Distributional structure. Word, 10(2-3), 146-162. doi:10.1080/00437956.1954.11659520
[25] Firth, J. R. (1957). A synopsis of linguistic theory 1930-55. 1952-59, 1-32.
[26] Turian, J., Ratinov, L. & Bengio, Y. (2010). Word representations: a simple and general method for semi-supervised learning. Proceedings of the 48th annual meeting of the association for computational linguistics (ACL), 384–394.
[27] Chen, S. C., Hung, H. T., Chen, B. & Chen, K. Y. (2015). Exploring Word Embedding and Concept Information for Language Model Adaptation in Mandarin Large Vocabulary Continuous Speech Recognition. Proceedings of the 27th Conference on Computational Linguistics and Speech Processing, 4-17.
[28] Hinton, G. E. (1986). Learning distributed representations of concepts. Proceedings of the Eighth Annual Conference of the Cognitive Science Society, 1–12.
[29] Mikolov, T., Chen, K., Corrado, G. S., & Dean, J. (2013). Efficient Estimation of Word Representations in Vector Space. CoRR, abs/1301.3781.
[30] Rumelhart, D.E., Hinton, G. E. & Williams, R. J. (1986). Learning Internal Representation by Error Propagation. Parallel Distributed Processing: Explorations in the Microstructure of Cognition, 1, 318-362.
[31] Olah, C. (2015). Understanding LSTM Networks. Retrieved from http://colah.github.io/posts/2015-08-Understanding-LSTMs/
[32] Britz, D. (2015). Recurrent Neural Networks Tutorial: Part 3 – Backpropagation Through Time and Vanishing Gradients. Retrieved from
http://www.wildml.com/2015/10/recurrent-neural-networks-tutorial-part-3-backpropagation-through-time-and-vanishing-gradients/
[33] Hochreiter, S. & Schmidhuber, J. (1997). Long short-term memory, Neural computation, 9(8), 1735-1780.
[34] Friedman, J. H. (1991). Multivariate Adaptive Regression Splines, The Annals of Statistics, 19(1), 1-67.
[35] Palshikar, G. K. (2007). Keyword extraction from a single document using centrality measures. Proceedings of the International Conference on Pattern Recognition and Machine Intelligence, 503-510.