Stock market prediction using Twitter

• Název práce v češtině: Využití nálady na Twitteru k predikcím trhu
• Název v anglickém jazyce: Stock market prediction using Twitter
• Klíčová slova anglicky: sentiment twitter market prediction
• Akademický rok vypsání: 2016/2017
• Typ práce: bakalářská práce
• Jazyk práce: angličtina
• Ústav: Institut ekonomických studií (23-IES)
• Vedoucí / školitel: prof. PhDr. Ladislav Krištoufek, Ph.D.
• Řešitel: skrytý - zadáno vedoucím/školitelem
• Datum přihlášení: 27.01.2017
• Datum zadání: 27.01.2017
• Datum a čas obhajoby: 13.06.2017 09:00
• Místo konání obhajoby: Opletalova - Opletalova 26, O105, Opletalova - místn. č. 105
• Datum odevzdání elektronické podoby: 19.05.2017
• Datum proběhlé obhajoby: 13.06.2017
• Oponenti: Mgr. Tomáš Křehlík, Ph.D.

Seznam odborné literatury

Bollen, Johan, and Huina Mao. 2011. “Twitter Mood As A Stock Market Predictor”. Computer 44 (10):
91-94. doi:10.1109/MC.2011.323.
Rexha, Andi, Mark Kröll, Mauro Dragoni, and Roman Kern. “Polarity Classification For Target Phrases
In Tweets: A Word2Vec Approach”, 217. doi:10.1007/978-3-319-47602-5_40.
MARTÍNEZ-CÁMARA, EUGENIO, M. TERESA MARTÍN-VALDIVIA, L. ALFONSO UREÑALÓPEZ,
and A RTURO MONTEJO-RÁEZ. 2014. “Sentiment Analysis In Twitter”. Natural Language
Engineering 20 (01): 1-28. doi:10.1017/S1351324912000332.
Makrehchi, Masoud, Sameena Shah, and Wenhui Liao. 2013. “Stock Prediction Using Event-Based
Sentiment Analysis”. 2013 Ieee/wic/acm International Joint Conferences On Web Intelligence (Wi) And
Intelligent Agent Technologies (Iat). IEEE, 337-342. doi:10.1109/WI-IAT.2013.48.
Zhang, Xue, Hauke Fuehres, and Peter A. Gloor. 2011. “Predicting Stock Market Indicators Through
Twitter “I Hope It Is Not As Bad As I Fear””. Procedia - Social And Behavioral Sciences 26: 55-62.
doi:10.1016/j.sbspro.2011.10.562.

Předběžná náplň práce v anglickém jazyce

Research question and motivation
Is it possible to predict market movements using Twitter sentiment?

The influence of the social media on the current society is one of the biggest game-changers in trading. Even though efficient
market hypothesis states that “stock market prices are largely driven by new information and follow a random walk pattern”,
it seems evident that a single tweet can cause to move a specific stock go up or down. Such can be the case the Twitter itself
– when their disappointing results leaked online, its stocks decreased by 20%. And as the whole Twitter is composed from
information bits like this, it seems intuitive that if we analyze all of the tweets it might be possible to predict the movement
of the whole market.
Methodology
We are going to use publicly available SNAP Twitter dataset consisting of 476 million tweets from period between June to
December 2009. We will analyze every tweet and and assign him value of its overall sentiment.
The widely spread method in the sentiment analysis research is lexicon based sentiment analysis. The main essence of this
method is lexicon of emotionally tinged words, and an algorithm looking at every sentence in a given corpus. Even though
this method is generally effective, it has its drawbacks. If we consider word such as ‘death’, it is generally considered that it
reflects negative sentiment. But in the case of one-time events such as death of Osama bin Laden that might not be true and
therefore our results could be biased on such days.
I would like to use small lexicon of so called “anchor words” – words unambiguously expressing certain emotion only. We
will obtain the words from Profile of Mood States (POMS) questionnaire. Then, using word2vec algorithm, we will obtain
similar words to every given anchor word for every day.
Using these, I will obtain the sentiment of the given tweet on given day. I will sum the obtained sentiment. In the end, I will
regress DJIA on the obtained sentiment from the previous day.

Contribution
Even though there is a lot of companies focused on sentiment trading, there has been done only little research. We can
evaluate whether these methods are capable of predicting market movements.
Other contribution is the new method of sentiment analysis. This method should theoretically be robust to unexpected events
and therefore this method could have more precise results.
Outline
1) Introduction
2) Literature review
3) Explaining word2vec algorithm
4) Exploratory dataset analysis
5) Sentiment analysis
6) Model learning
7) Prediction of Dow Jones Industrial Average
8) Discussion of results
9) Further work
10) Literature