Please use this identifier to cite or link to this item: https://ah.nccu.edu.tw/handle/140.119/118938


Title: 以深度學習實作情感分析於台灣股價指數趨勢之研究—以光電類股為例
Applying deep learning to sentiment analysis for prediction of Taiwan Optoelectronic Index trend
Authors: 黃彥霖
Huang, Yen Lin
Contributors: 姜國輝
季延平

黃彥霖
Huang, Yen Lin
Keywords: 情感分析
LDA主題模型
支援向量機
長短期記憶網路
Sentimental analysis
LDA
SVM
LSTM
Date: 2018
Issue Date: 2018-07-27 11:40:24 (UTC+8)
Abstract: 光電是一門結合光學、電子與電機的技術,以光電技術為核心,合成各類零組件、設備,並廣泛應用於通訊、生化、醫療、工業、能源等各種領域。近年來台灣的精密光學元件發展快速,受惠於智慧型手機興起,帶動了台灣光電產業發展。據過往研究指出,網路上的文本訊息會對群眾情緒造成影響,進而影響股價波動,因此若能將網路上大量財經文本快速分析,來推測投資大眾情緒,進而預測走勢,可以幫助投資者增加其獲利。為檢驗新聞情緒對於價格預測的重要性,本研究嘗試使用混和非監督式學習與監督式學習的方式進行實驗模型的建立。

在非監督式學習方面,本研究利用LDA主題模型,將光電類股之財經新聞文本,以進行主題分群。將各主題之文本,藉由比對情感詞集的方式進行情緒指數之計算,藉此找到各文本的情緒傾向。接著利用視覺化的方式,找出對後續分類模型有所助益的領先指標特性之主題。

在監督式學習方面,本研究分別建立SVM分類模型與LSTM神經網路分類模型,針對光電類股的財經新聞文本進行情感分析,並與單純使用技術指標之分類模型做比較。實驗結果發現,結合技術指標與情緒指數之準確率,SVM模型約為70%,神經網路模型約為60%。在股價漲跌分類方面,SVM模型優於LSTM分類模型,但不論是使用SVM模型或是LSTM模型,混和技術指標與情緒指數之模型在準確率上都高於單純只用技術指標之模型,前者較後者可提升多達7%之準確率,顯示情感分析確實能有效提升光電類股指數趨勢模型之準確度。
The optoelectronic is a technology that combines optics and electronics. In recent years, Taiwan's precision optical components have developed rapidly, benefiting from the rise of smart phones, which has driven the development of Taiwan's optoelectronic industry. According to past studies, text documents on the Internet will affect the mood of the investors, and then affect stock price indirectly. For investors, it is important to know how to analyze the potential emotion in text documents and use it to predict the stock trend. In order to test the importance of text emotions for price prediction, this research established a model that mixed unsupervised learning and supervised learning.

In the part of unsupervised learning, this research used the LDA model to assign documents to topics. Next, calculated the sentiment index to find the sentiment tendency of each documents. Then this research used the visualization to find the topic that was probably a leading indicator and helpful to classification model.

In the part of supervised learning, this research established the SVM model and the LSTM neural network model respectively. The result showed that, mixed with technical indicators and sentiment index, the accuracy of the SVM model was about 70%, and that of the neural network model was about 60%. In classification, the SVM model was better than the LSTM classification model, but in both of the SVM model and the LSTM model, the accuracy of the model that included the sentiment index and technical index was higher than that which included only technical index. The former could improve the accuracy by up to 7% compared with the latter, showing that sentiment analysis was able to improve the accuracy of the prediction of Optoelectronics stock index trend effectively.
Reference: [ 1 ] 王正豪, & 李啟菁. (2010). 中文部落格文章之意見分析. 國立台北科技大學資訊工程研究所碩士論文,
[ 2 ] 邱世芳(2008)。台灣地區光電產業之廠商衍生與空間擴散。成功大學都市計劃學系碩士論文
[ 3 ] 李啟菁. (2010). 中文部落格文章之意見分析. 臺北科技大學資訊工程系研究所學位論文, 1-44.
[ 4 ] 林育龍. (2014). 對使用者評論之情感分析研究-以 Google Play 市集為例
[ 5 ] 洪崇洋. (2012). LDA 和使用紀錄為基礎的線上電子書主題趨勢發掘方法. 國立中山大學資訊管理所碩士論文
[ 6 ] 張日威. (2014). 應用LDA進行Plurk主題分類及使用者情緒分析. 國立雲林科技大學資訊管理所碩士論文
[ 7 ] 雷祖強, 周天穎, 萬絢, 楊龍士, & 許晉嘉. (2007). 空間特徵分類器支援向量機之研究. Journal of Photogrammetry and Remote Sensing, 12(2), 145-163.
[ 8 ] 劉吉軒, & 吳建良. (2007). 以情緒為中心之情境資訊觀察與評估. Paper presented at the NCS全國計算機會議.
[ 9 ] 劉羿廷. (2015). 運用財經文本情感分析於台灣電子類股價指數趨勢預測之研究
[ 10 ] 蕭昱維. (2014). 基於多階 LDA 技術尋找 Twitter 文章的隱含主題之研究. 樹德科技大學資訊工程系碩士班學位論文, 1-47.
[ 11 ] Blei, D. M., Ng, A. Y., & Jordan, M. I. (2003). Latent dirichlet allocation. Journal of machine Learning research, 3(Jan), 993-1022.
[ 12 ] Boser, B. E., Guyon, I. M., & Vapnik, V. N. (1992, July). A training algorithm for optimal margin classifiers. In Proceedings of the fifth annual workshop on Computational learning theory (pp. 144-152). ACM.
[ 13 ] Griffiths, T. L., & Steyvers, M. (2004). Finding scientific topics. Proceedings of the National academy of Sciences, 101, 5228-5235.
[ 14 ] Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
[ 15 ] Liu, B. (2012). Sentiment analysis and opinion mining. Synthesis lectures on human language technologies, 5(1), 1-167.
[ 16 ] Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781.
[ 17 ] McCulloch, W. S., & Pitts, W. (1943). A logical calculus of the ideas immanent in nervous activity. The bulletin of mathematical biophysics, 5(4), 115-133.
[ 18 ] Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1986). Learning representations by back-propagating errors. nature, 323(6088), 533.
Description: 碩士
國立政治大學
資訊管理學系
105356035
Source URI: http://thesis.lib.nccu.edu.tw/record/#G0105356035
Data Type: thesis
Appears in Collections:[資訊管理學系] 學位論文

Files in This Item:

File SizeFormat
603501.pdf2288KbAdobe PDF0View/Open


All items in 學術集成 are protected by copyright, with all rights reserved.


社群 sharing