Exploiting Social Relations and Sentiment for Stock Prediction

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1139–1145, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Exploiting Social Relations and Sentiment for Stock Prediction Jianfeng Si* Arjun Mukherjee† Bing Liu† Sinno Jialin Pan* Qing Li‡ Huayi Li† * Institute for Infocomm Research, Singapore { thankjeff@gmail.com, jspan@i2r.a-star.edu.sg} †Department of Computer Science, University of Illinois at Chicago, Chicago, IL 60607, USA { arjun4787@gmail.com, liub@cs.uic.edu, lhymvp@gmail.com} ‡ Department of Computer Science, City University of Hong Kong, Hong Kong, China qing.li@cityu.edu.hk Abstract In this paper we first exploit cash-tags (“$” fol- lowed by stocks’ ticker symbols) in Twitter to build a stock network, where nodes are stocks connected by edges when two stocks co-occur frequently in tweets. We then employ a labeled topic model to jointly model both the tweets and the network structure to assign each node and each edge a topic respectively. This Semantic Stock Network (SSN) summarizes discussion topics about stocks and stock relations. We fur- ther show that social sentiment about stock (node) topics and stock relationship (edge) topics are predictive of each stock’s market. For predic- tion, we propose to regress the topic-sentiment time-series and the stock’s price time series. Ex- perimental results demonstrate that topic senti- ments from close neighbors are able to help im- prove the prediction of a stock markedly. 1 Introduction Existing research has shown the usefulness of public sentiment in social media across a wide range of applications. Several works showed so- cial media as a promising tool for stock market prediction (Bollen et al., 2011; Ruiz et al., 2012; Si et al., 2013). However, the semantic relation- ships between stocks have not yet been explored. In this paper, we show that the latent semantic relations among stocks and the associated social sentiment can yield a better prediction model. On Twitter, cash-tags (e.g., $aapl for Apple Inc.) are used in a tweet to indicate that the tweet talks about the stocks or some other related in- formation about the companies. For example, one tweet containing cash-tags: $aapl and $goog (Google Inc.), is “$AAPL is loosing customers. everybody is buying android phones! $GOOG”. Such joint mentions directly reflect some kind of latent relationship between the involved stocks, which motivates us to exploit such information for the stock prediction. We propose a notion of Semantic Stock Net- work (SSN) and use it to summarize the latent semantics of stocks from social discussions. To our knowledge, this is the first work that uses cash-tags in Twitter for mining stock semantic relations. Our stock network is constructed based on the co-occurrences of cash-tags in tweets. With the SSN, we employ a labeled topic model to jointly model both the tweets and the network structure to assign each node and each edge a topic respectively. Then, a lexicon-based senti- ment analysis method is used to compute a sen- timent score for each node and each edge topic. To predict each stock’s performance (i.e., the up/down movement of the stock’s closing …

Content truncated to 500 words