Learning to Generate Market Comments from Stock Prices

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1374–1384 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1126 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1374–1384 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1126 Learning to Generate Market Comments from Stock Prices Soichiro Murakami†,∗ Akihiko Watanabe †,∗ Akira Miyazawa ‡,¶,∗ Keiichi Goshima †,∗ Toshihiko Yanase § Hiroya Takamura †,∗ Yusuke Miyao ‡,¶,∗ † Tokyo Institute of Technology ‡ The Graduate University for Advanced Studies § Hitachi, Ltd. ¶ National Institute of Informatics ∗ National Institute of Advanced Industrial Science and Technology {murakami,watanabe}@lr.pi.titech.ac.jp, goshima.k.aa@trn.dis.titech.ac.jp, takamura@pi.titech.ac.jp toshihiko.yanase.gm@hitachi.com, {miyazawa-a,yusuke}@nii.ac.jp Abstract This paper presents a novel encoder- decoder model for automatically generat- ing market comments from stock prices. The model first encodes both short- and long-term series of stock prices so that it can mention short- and long-term changes in stock prices. In the decoding phase, our model can also generate a numerical value by selecting an appropriate arithmetic op- eration such as subtraction or rounding, and applying it to the input stock prices. Empirical experiments show that our best model generates market comments at the fluency and the informativeness approach- ing human-generated reference texts. 1 Introduction Various industries such as finance, pharmaceuti- cals, and telecommunications have been increas- ingly providing opportunities to treat various types of large-scale numerical time-series data. Such data are hard for non-specialists to interpret in de- tail and time-consuming even for specialists to con- strue. As a result, there has been a growing interest in automatically generating concise descriptions of such data, i.e., data summarization. This interest in data summarization is encouraged by the recent de- velopment of neural network-based text generation methods. Given an appropriate architecture, a neu- ral network can generate a sentence that is mostly grammatical and semantically reasonable. In this study, we focus on the task of gen- erating market comments from a time-series of stock prices. We adopt an encoder-decoder model (Sutskever et al., 2014) and exploit its capability to learn to capture the behavior of the input and generate a description of it. Although encoder- decoder models can learn to do this, they need to be (1) (2) (3) (4) (5) (6) Previous Day (Afternoon Session) Morning Session Afternoon Session 19200 19300 19400 19500 19600 14:00 15:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 Time Stock price [yen] Time Comment (1) 09:00 Nikkei opens with a continual fall. (2) 09:29 Nikkei turns to rise. (3) 11:30 Nikkei continues to fall. The clos- ing price of the morning session de- creases by 5 yen to 19,386 yen. (4) 12:30 Nikkei rises at the beginning of the afternoon session. (5) 13:54 Nikkei gains more than 100 yen. (6) 15:00 Nikkei rebounds and closes up 102 yen to 19,494 yen. Figure 1: Nikkei 225 and market comments. provided with an appropriate network-architecture and necessary information. We use Figure 1 to illustrate the characteristic problems …

Content truncated to 500 words