Incorporating Fine-grained Events in Stock Movement Prediction

Proceedings of the Second Workshop on Economics and Natural Language Processing, pages 31–40 Hong Kong, November 4. c⃝2019 Association for Computational Linguistics 31 Incorporating Fine-grained Events in Stock Movement Prediction Deli Chen1∗, Yanyan Zou2, Keiko Harimoto3, Ruihan Bao3, Xuancheng Ren1, Xu Sun1 1MOE Key Lab of Computational Linguistics, School of EECS, Peking University 2StatNLP Research Group, Singapore University of Technology and Design 3Mizuho Securities Co., Ltd. {chendeli,renxc,xusun}@pku.edu.cn, yanyan zou@mymail.sutd.edu.sg, {keiko.harimoto,ruihan.bao}@mizuho-sc.com Abstract Considering event structure information has proven helpful in text-based stock movement prediction. However, existing works mainly adopt the coarse-grained events, which loses the specific semantic information of diverse event types. In this work, we propose to incor- porate the fine-grained events in stock move- ment prediction. Firstly, we propose a pro- fessional finance event dictionary built by do- main experts and use it to extract fine-grained events automatically from finance news. Then we design a neural model to combine finance news with fine-grained event structure and stock trade data to predict the stock move- ment. Besides, in order to improve the gener- alizability of the proposed method, we design an advanced model that uses the extracted fine- grained events as the distant supervised label to train a multi-task framework of event ex- traction and stock prediction. The experimen- tal results show that our method outperforms all the baselines and has good generalizability. 1 Introduction Stock movement plays an important role in eco- nomic activities, so the prediction of stock move- ment has caught a lot of attention of researchers. In recent years, employing the stock related text (such as finance news or tweets) has become the mainstream (Si et al., 2014; Ding et al., 2015; Li et al., 2015; Alostad and Davulcu, 2017; Zhong et al., 2017; Zhang et al., 2018a) of stock move- ment prediction task. In these text-based stock prediction works, various methods are proposed to extract semantic information from stock re- lated text to help the prediction of stock move- ment. There are mainly two methods of applying text: employing raw text (Hu et al., 2018; Xu and Cohen, 2018) or coarse-grained <S,P,O> struc- ture (subject, predicate and object) extracted from ∗This work is done when Deli Chen is a intern at Mizuho Securities. • Daiseki 9-mth group recurring profit is 5.61 bln yen (-11.4 %). Raw Text • Subject: group recurring -----,------profit • Predicate: is • Object: 5.61 bln yen Coarse-grained Event • Firm: Daiseki • Type: group recurring • Value: 5.61 bln yen • Change Rate: -11.4 % • Time: 9-mth Fine-grained Event Figure 1: The same news of Earnings Profit event in different forms. The event structure consists of event roles (red words) which are the key point of the seman- tic information. text (Ding et al., 2016; Zhang et al., 2018b). In the previous studies, the latter method has proven more powerful than the former one, which demon- strates that the event structure containing semantic information is helpful for stock movement predic- tion. Figure 1 shows a piece of news of Earn- …

Content truncated to 500 words