NLG301 at SemEval-2017 Task 5: Fine-Grained Sentiment Analysis on Financial Microblogs and News

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 847–851, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics NLG301 at SemEval-2017 Task 5: Fine-Grained Sentiment Analysis on Financial Microblogs and News Chung-Chi Chen, Hen-Hsen Huang and Hsin-Hsi Chen Department of Computer Science and Information Engineering National Taiwan University, Taipei, Taiwan {cjchen, hhhuang}@nlg.csie.ntu.edu.tw; hhchen@ntu.edu.tw Abstract Short length, multi-targets, target relation- ship, monetary expressions, and outside reference are characteristics of financial tweets. This paper proposes methods to extract target spans from a tweet and its referencing web page. Total 15 publicly available sentiment dictionaries and one sentiment dictionary constructed from training set, containing sentiment scores in binary or real numbers, are used to com- pute the sentiment scores of text spans. Moreover, the correlation coefficients of the price return between any two stocks are learned with the price data from Bloomberg. They are used to capture the relationships between the interesting tar- get and other stocks mentioned in a tweet. The best result of our method in both sub- task are 56.68% and 55.43%, evaluated by evaluation method 2. 1. Introduction Nowadays the discussion of finance on social media such as twitter reveals the feeling of a mar- ket in some degrees. Market sentiment analysis becomes an important financial technology in trading strategies (Kazemian, 2014). Financial tweet sentiment classification differs from tradi- tional sentiment classification in several ways. Firstly, the sentiment degree is a real number ra- ther than discrete numbers such as 1 (positive), 0 (neutral), and -1 (negative). Secondly, a financial tweet usually concerns multiple targets. In the tweet, “Oil To Break Out: Adding Chevron https://t.co/IrZkAVxjiE $AXP $CLGRF $CSCO $ERX $IBM $MCD $SSRI $VLO $WMT $XOM $CVX”, all companies de- noted by the cashtag $ticker-symbol share only one description. The activity, oil to break out, is a good news for energy companies, but may be bad for shipping companies. Domain knowledge about the targets is necessary for suitable interpre- tation in this case. Thirdly, numbers in the financial tweets are quite important. In the tweet, “MarketWatch: RT wmwitkowski: Guess who sold off about $800 million in $MDLZ after losing about $1 billion on $VRX???https://t.co/SHiJutyenv”, large number means more negative. In contrast, in the tweet (named T1 hereafter), “$AAPL now up 2.2% w/div since my original call, while $SPY up only 0.6% even w/ this Fri's div. #EMH be damned. Still holding”, the larger the number is, the more positive score the target will get. The activity re- lated to numbers determines the polarity and its degree. Fourthly, sentiment scores depend on the activ- ity of the companies, and their relationships, e.g., the adversarial relation versus the cooperate rela- tion. In the tweet, “Report: Apple signs up for Google's cloud, uses much less of Amazon's $AAPL $GOOG $GOOGL $AMZN $DROPB https://t.co/zN3KDGYvGT”, $AAPL $GOOGL $AMZN and $DROPB are assigned sentiment scores 0.15, 0.443, -0.38 and -0.213, respectively, by human annotators because Amazon and Drop- box are two competitors of Google in the cloud market. This paper …

Content truncated to 500 words