以專利探勘與財務新聞情緒分析建立企業績效評估模型之研究

Developing a Firm Performance Evaluating Model based on Patent Mining and Financial News Sentiment Analysis

張培育(Jhang Pei-Yu)

指導教授 ： 陳灯能 ； 賴佳瑜

國立屏東科技大學/管理學院/資訊管理系所/碩士(2020年)

https://doi.org/10.6346/NPUST202000314

摘要

隨著資訊科技蓬勃發展，取得有關企業動向的資訊更加容易，然而，預測企業未來之財務績效仍然是一個極具挑戰性且重要的任務，如何有效地預測企業未來業績將關係到企業管理人員、投資人之決策判斷，甚至是產業及市場發展。近年來，保護智慧產權之意識逐漸抬頭，企業投入大量資源於研發生產，專利資源隨之迅速增長成為富有研究價值之重要資源，企業所持有之專利資源也因此成為衡量企業獲益能力之重要指標。除企業自身之經營能力外，新聞輿論與產業動向亦是反應企業獲益情況，因此將自然語言處理(NLP)與機器學習技術應用至財務金融新聞以協助財務分析也成為目前熱門之議題。因此，本研究結合專利探勘、情緒分析技術、財務指標等構面，並以隨機森林、XGBoost及GradientBoost等作為分類器，評估結合三個構面之模型。本研究以「蘋果日報」網路新聞做為財務新聞之來源及自「臺灣國際專利局」獲取之專利資源，結合臺灣上市企業之財務季度報表評估本研究所提出之方法，且實證結果顯示該方法能提供較高之準確率。

關鍵字

專利探勘 ； 情緒分析 ； 隨機森林 ； XGBoost ； GradientBoost

並列摘要

With the rapid development of information technology, it is easier to obtain information about the enterprise’s development trends. However, predicting the future financial performance of the enterprise is still a very challenging and essential task. How to effectively predict the future performance of the enterprise will be related to enterprise managers, investor’s decision-making, even industry development. In recent years, the awareness of protecting intellectual property rights has gradually risen. Enterprises have invested a lot of resources in R&D and production, and patent resources have rapidly grown to become essential resources to predict the profitability of enterprises. In addition to the enterprise’s own operational capabilities, public opinion news and industry trends also reflect the enterprise’s profitability. Therefore, using Natural Language Processing (NLP) and machine learning technology to extract useful information from the finance news and other textual resources for financial analysis also becomes a popular topic. In this research, we combine three major components of patent mining, aspect-based sentiment analysis, and financial indicators technologies and apply the SVM, Random Forest, XGBoost, and GradientBoost as the classifiers to evaluate the model of predicting firm performance. We further adapt the feature importance analysis to identify significant indicators for the prediction model. In this study, we obtain financial news from the Apple Daily News website and patent data from the Taiwan International Patent Office, combining the quarterly reports of listed companies in the Taiwan market to evaluate our proposed model. With the experiments, the proposed model has been proved that it can provide higher accuracy in predicting firm performance.

並列關鍵字

Patent Mining ； Sentiment Analysis ； Random Forest ； GradientBoost ； XGBoost

參考文獻

Abbas, A., Zhang, L., and Khan, S. U. 2014. "A Literature Review on the State-of-the-Art in Patent Analysis," World Patent Information (37), pp. 3-13.

Allahyari, M., Pouriyeh, S., Assefi, M., Safaei, S., Trippe, E. D., Gutierrez, J. B., and Kochut, K. J. a. p. a. 2017. "A Brief Survey of Text Mining: Classification, Clustering and Extraction Techniques," arXiv preprint arXiv:1707.02919).

Allozi, N. M., and Obeidat, G. S. J. J. o. S. S. 2016. "The Relationship between the Stock Return and Financial Indicators (Profitability, Leverage): An Empirical Study on Manufacturing Companies Listed in Amman Stock Exchange," (5:3), pp. 408-424.

Aristodemou, L., and Tietze, F. 2018. "The State-of-the-Art on Intellectual Property Analytics (Ipa): A Literature Review on Artificial Intelligence, Machine Learning and Deep Learning Methods for Analysing Intellectual Property (Ip) Data," World Patent Information (55), pp. 37-51.

Bagheri, A., Saraee, M., and de Jong, F. 2013. "Care More About Customers: Unsupervised Domain-Independent Aspect Detection for Sentiment Analysis of Customer Reviews," Knowledge-Based Systems (52), pp. 201-213.