Financial Keyword Expansion via Continuous Word Vector Representations

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1453–1458, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Financial Keyword Expansion via Continuous Word Vector Representations Ming-Feng Tsai Department of Computer Science & Program in Digital Content and Technology National Chengchi University Taipei 116, Taiwan mftsai@nccu.edu.tw Chuan-Ju Wang Department of Computer Science University of Taipei Taipei 100, Taiwan cjwang@utaipei.edu.tw Abstract This paper proposes to apply the contin- uous vector representations of words for discovering keywords from a financial sen- timent lexicon. In order to capture more keywords, we also incorporate syntactic information into the Continuous Bag-of- Words (CBOW) model. Experimental re- sults on a task of financial risk prediction using the discovered keywords demonstrate that the proposed approach is good at pre- dicting financial risk. 1 Introduction In the present environment with a great deal of information, how to discover useful insights for decision making is becoming increasingly impor- tant. In finance, there are typically two kinds of information (Petersen, 2004): soft information usu- ally refers to text, including opinions, ideas, and market commentary, whereas hard information is recorded as numbers, such as financial measures and historical prices. Most financial studies related to risk analysis are based on hard information, es- pecially on time series modeling (Christoffersen and Diebold, 2000; Lee and Tong, 2011; Wu et al., 2014; Y¨uml¨u et al., 2005). Despite of using only hard information, some literature incorporates soft textual information to predict financial risk (Ko- gan et al., 2009; Leidner and Schilder, 2010; Tsai and Wang, 2013). Moreover, sentiment analysis, a technique to make an assessment of the sentiments expressed in various information, has also been applied to analyze the soft textual information in financial news, reports, and social media data (De- vitt and Ahmad, 2007; Loughran and McDonald, 2011; Wang et al., 2013). Continuous vector space models (Bengio et al., 2003; Schwenk, 2007; Mikolov et al., 2010) are neural network language models, in which words are represented as high dimensional real val- ued vectors. These representations have recently demonstrated promising results across variety of tasks (Schwenk, 2007; Collobert and Weston, 2008; Glorot et al., 2011; Socher et al., 2011; Weston et al., 2011), because of their superiority of capturing syntactic and semantic regularities in language. In this paper, we apply the Continuous Bag-of-Words (CBOW) model (Mikolov et al., 2013) on the soft textual information in financial reports for discov- ering keywords via financial sentiments. In spe- cific, we use the continuous vector representations of words to find out similar terms based on their contexts. Additionally, we propose a straightfor- ward approach to incorporate syntactic information into the CBOW model for better locating similarly meaningful or highly correlated words. To the best of our knowledge, this is the first work to incorpo- rate more syntactic information by adding Part-Of- Speech (POS) tags to the words before training the CBOW model. In our experiments, the corpora are the annual SEC1-mandated financial reports, and there are 3,911 financial sentiment keywords for …

Content truncated to 500 words