=Paper=
{{Paper
|id=Vol-1388/DeCat2015-paper3
|storemode=property
|title=Bitcoin Spread Prediction Using Social and Web Search Media
|pdfUrl=https://ceur-ws.org/Vol-1388/DeCat2015-paper3.pdf
|volume=Vol-1388
|dblpUrl=https://dblp.org/rec/conf/um/MattaLM15
}}
==Bitcoin Spread Prediction Using Social and Web Search Media==
https://ceur-ws.org/Vol-1388/DeCat2015-paper3.pdf
Bitcoin Spread Prediction Using Social And Web
Search Media
Martina Matta, Ilaria Lunesu, Michele Marchesi
Università degli Studi di Cagliari
Piazza d’Armi, 09123 Cagliari, Italy
{martina.matta,ilaria.lunesu,michele}@diee.unica.it
Abstract. In the last decade, Web 2.0 services such as blogs, tweets,
forums, chats, email etc. have been widely used as communication me-
dia, with very good results. Sharing knowledge is an important part of
learning and enhancing skills. Furthermore, emotions may affect decision-
making and individual behavior. Bitcoin, a decentralized electronic cur-
rency system, represents a radical change in financial systems, attracting
a large number of users and a lot of media attention. In this work, we
investigated if the spread of the Bitcoin’s price is related to the volumes
of tweets or Web Search media results. We compared trends of price with
Google Trends data, volume of tweets and particularly with those that
express a positive sentiment. We found significant cross correlation val-
ues, especially between Bitcoin price and Google Trends data, arguing
our initial idea based on studies about trends in stock and goods market.
Keywords: Bitcoin, Twitter, Google Trends, Sentiment Analysis
1 Introduction
Bitcoin, a decentralized electronic currency system, represents a radical change
in financial systems after its creation in 2008 by Satoshi Nakamoto [22]. Bitcoin
stands for an IT innovation based on the advancement in peer-to-peer networks
[21] and cryptographic protocols. Due to its properties, Bitcoin is not managed
by any governments or bank. Like any other currency, a peculiarity of Bitcoin
is to facilitate transactions of services and goods [20], attracting a large number
of users and a lot of media attention.
Nowadays, Web 2.0 services such as blogs, tweets, forums, chats, email etc.
are widely used as communication media, with satisfying results. Sharing knowl-
edge is an important part of learning and enhancing skills. Through the use
of social media services, team members have the opportunity to acquire more
detailed information about their peers’ expertise [7]. Social media data repre-
sents a collective indicator of thoughts and ideas regarding every aspect of the
world. It has been possible to assist to deep changes in habits of people in the
use of social media and social network. Twitter[10], an online social networking
website and microblogging service, has become an important tool for businesses
and individuals to communicate and share information with a rapid growth and
significant adoption. In addition, Twitter has rapidly grown as a mean to share
ideas and thoughts on investing decisions.
In this work we analyze whether social media activity or information ex-
tracted by web search media could be helpful and used by investment profes-
sionals. There are several works that present predictive relationships between
social media and bitcoin price where the relative effects of different social media
platforms (Internet forum vs. microblogging) and the dynamics of the resulting
relationships, are analyzed using cross-correlation such as [17] or linear regression
analysis such as [6] or [5]. Social factors, that are composed of interactions among
the actors of the market, may strongly drive dynamics of Bitcoin’s economy [3].
We decided to apply automated Sentiment Analysis on shared short messages
of users on Twitter in order to automatically analyze people’s opinions, senti-
ments, evaluations and attitudes. We investigated whether public sentiment, as
expressed in large-scale collections of daily Twitter posts, can be used to predict
the Bitcoin market. We tried to discover if the chatter of the community can be
used to make qualitative predictions about Bitcoin market, attempting to estab-
lish whether there is any correlation between tweet’s sentiment and the Bitcoin’s
price 1 . The results suggest that a significant relationship with future Bitcoin’s
price and volume of tweets exists on a daily level. We also used Google Trends to
analyze Bitcoin’s popularity under the perspective of Web search, which provides
a time series index of the volume of queries made by users in Google Search. We
found a striking correlation between Bitcoin’s price spread and changes in query
volumes for the “Bitcoin” search term.
The body of this paper is organized in five major sections. Section 2, describes
the background, section 3 presents the research steps of our study and section
4 summarizes and discusses our results. Finally, section 5 presents conclusions
and suggestions for future work.
2 Background
In these decades, social web has been commercially exploited for goals such as
automatically extracting customer opinions about products or brands, to find
which aspects are liked and which are disliked [9]. In their work, Ye and Wu
demonstrate how particularly interesting is the influence of Twitter users and
the propagation of the information related to their tweets[18].
According to Alexa [12], Twitter had become the world’s seventh most pop-
ular website by March 2015. Twitter [10] is an online social networking website
and microblogging service that allows users to post and read text-based mes-
sages of up to 140 characters, known as "tweets". Launched in July of 2006 by
Jack Dorsey, Twitter is now in the top 10 most visited internet sites with a total
amount of 645,750,000 registered users. Java et al. affirm that it seems to be
used to share information and to describe minor daily activities [14]. The short
format of a tweet is a defined characteristic of the service, allowing informal col-
laboration and quick information sharing. For business, Twitter can be used to
1
https://markets.blockchain.info/
broadcast company’s latest news, posts, read comments of the customers or in-
teract with them. A communicative feature of Twitter is the hashtag: a metatag
beginning with the character #, designed to help others find a post.
Twitter is a rich source of real-time information regarding current societal trends
and opinions. There are also studies that report another use of Twitter, namely
as a possible predictor of market trends. Indeed, in 2010, a publication of the
professor Johan Bollen showed that combining information on Wall Street with
the millions of Tweets and posts, makes possible to anticipate financial perfor-
mance [6]. In this work, Granger causality analysis and a Self-Organizing Fuzzy
Neural Network are used to investigate the hypothesis that public mood states,
as measured by the OpinionFinder and GPOMS mood time series, are predictive
of changes in DJIA closing values. The analysis of Tweets made by Bollen would
have had 87% of chance to successfully predict prices of the stock, 3 or 4 days
in advance. This study and analysis of millions of posts on Twitter represents a
thermometer of emotions, on a large scale, which reflects the whole of society.
Earlier studies had found that blogs can be used to evaluate public mood, and
that tweets about movies can predict box office sales. Investigating the literature
related to different uses of social media, and Twitter in particular, we collected
information about the use of Twitter for seeking real world emotions that could
predict real financial markets trend [1]. In their paper, Rao and Srivastava inves-
tigate the complex relationship between tweet board literature (like bullishness,
volume, agreement etc) with the financial market instruments (like volatility,
trading volume and stock price) [2].
The Bitcoin represents an important new phenomenon in financial markets. Mai
et al. [4] examine predictive relationships between social media and Bitcoin re-
turns by considering the relative effect of different social media platforms (In-
ternet forum vs. microblogging) and the dynamics of the resulting relationships
using vector autoregressive and vector error correction models.
In their work, Garcia et al. [3] show the interdependence between social signals
and price in the Bitcoin economy, namely a social feedback cycle based on word-
of-mouth effect and a user-driven adoption cycle. They provide evidence that
Bitcoin’s growing popularity causes an increasing search volumes, which in turn
result a higher social media activity about Bitcoin. More interest inspire the
purchase of bitcoins by users, driving the prices up, which eventually feeds back
on the search volumes.
We compared Twitter’s trending topic about Bitcoin with those in other media,
namely, Google Trends [8]. This is a feature of Google search engine that illus-
trates how frequently a fixed search term was looked for. Through this, you can
compare up to five topics at one time to view relative popularity, allowing you
to gain an understanding of the hottest search trends of the moment, along with
those developing in popularity over time. Following this kind of approach, we
evaluated how much “bitcoin” term, for the analyzed time interval, is looked for
using Google’s search engine.
3 Methodology
3.1 Sentiment Analysis
Tweets sometimes express opinions about different topics, and for this reason we
decided to evaluate user’s opinion about Bitcoin. We also investigated its power
at predicting real-world outcomes. In order to evaluate if a user really appreciates
the Bitcoin spread, we tried to predict sentiments analyzing tweets collection. In
recent years, there is a wide collection of research surrounding machine learning
techniques, in order to extract and identify subjective information in texts. This
area is known as sentiment analysis or opinion mining [15]. Sentiment techniques
are able to extract indicators of public mood directly from social media content
[24].
Pang et al. argue that the research field of sentiment analysis has developed
many algorithms to identify if the opinion expressed is positive or negative. In
fact, algorithms to recognize sentiment are required to understand the role of
emotions in informal communications [15]. Go et al. affirmed the strength of
the sentiment analysis applied to the Twitter domain by using similar machine
learning techniques to classifying the sentiment of tweets [19].
We chose to use automated sentiment analysis techniques to identify the
sentiments of tweets in the matter of Bitcoin. Since the goal of this research
is neither to develop a new sentiment analysis nor to improve an existing one,
we used "SentiStrenght", a tool developed by a team of researchers in the UK
that demonstrated good outcomes [11]. SentiStrength estimates the strength
of positive and negative sentiments in short texts. It is based on a dictionary
of sentiment words, each one associated with a weight, which is its sentiment
strength. In addition, this method uses some rules for non-standard grammar.
Based on the formal evaluation of this system on a large sample of comments
from MySpace.com, the accuracy of predicting positive and negative emotions
was something similar to that of other systems (72.8% for negative emotions
and 60.6% for positive emotions, based on a scale of 1-5). Compared to other
methods, SentiStrenght showed the highest correlation with human coders [13].
The tool is able to assess each message separately and, at the end, it returns one
singular value: a positive, a negative or a neutral sentiment.
3.2 Data Collection
Tweets are available and are easily retrieved making use of Twitter Application
Programming Interface (API) [16]. Composing the hashtag #Bitcoin or @bit-
coin, we are able to gather all tweets that mentioned the analyzed subject. We
briefly describe the different components of our system. An overview of this
architecture is shown in Figure 1. The system consists of four components:
• Twitter Streaming API : it provides access to Twitter data, both public and
protected, on a nearly real-time basis. A persistent connection is created
between our system and Twitter. As soon as tweets come in, Twitter notifies
our system in real time, allowing us to store them into our database.
Fig. 1. System Architecture
• DataStore: our datastore consists of a back-end database engine, using MySQL
as RDBMS, that repeatedly saves the incoming tweets from the Twitter
Streaming API.
• SentiStrenght tool
• Java Module: this component allows us to send automated requests to Twit-
ter Streaming API, to recover new tweets about Bitcoin, to parse the data
gathered and to store them into our datastore. In a later stage, these data
are sent to SentiStrenght tool in order to automatically evaluate the users’
opinion.
We analyzed a collection of tweets, regarding Bitcoin, posted on Twitter between
January 2015 and March 2015 (60 days). During this time 1,924,891 tweets were
collected. The tweets were analyzed to determine its identifier, the date-time of
the submission, its type, and its text content, which is limited to 140 characters.
Comparing the timeline of tweets and the fluctuations in the Bitcoin market, we
determined the specific day that provide a better correlation value. We then used
SentiStrenght to evaluate comments extracted from Twitter. Given as input all
tweets, the system assigned a score for each comment:
• 1 if the comment is positive
• -1 if the comment is negative
• 0 if the comment is neutral
4 Results
In order to decide the correct strategy of analysis for studying the relationship
among Bitcoin’s price and others meaningful parameters, the available related
literature has been examined in depth. Most of articles [6] [1] [2] reports analysis
about the existent relationship between the volume of tweets and the market
evolution. In general, Bollen et al. demonstrated that tweets can predict the
market trend 3-4 days in advance, with a good chance of success. We analyzed
the Bitcoin price’s behavior comparing its variations with the number of tweets,
with the number of tweets with positive mood, and with Google Trends results.
The computation of cross-correlation yielded interesting results.
Our result seems to confirm that volumes of exchanged tweets may predict the
fluctuations of Bitcoin’s price. Furthermore, the comparison between tweets with
a positive mood and trend of Bitcoin’s price seems to prove this behavior. The
Fig. 2. Similarity between Bitcoin’s price and number of Tweets
examined literature shows different ways to highlight the existent relationship be-
tween big volumes of exchanged tweets and meaningful variations in the Bitcoin’s
price. Some papers show studies using regression methodology [4] or causality
analysis [6]. Rao et al.[2] and Mittal et al.[5] showed how goods and stocks mar-
kets may be influenced by a big exchanged of tweet’s volume. Inspired by these
works, we tried to demonstrate how chatter of tweets might predict the price’s
variations of Bitcoin. Figure 2 illustrates the curve trend of Bitcoin prices, ex-
pressed in dollars, and Twitter volume. We calculated the cross-correlation and,
analyzing the results, we found that, in minimal degree, tweets volume is related
to price with a maximum cross correlation value of 0.15 at a lag of 1 day (this
is not very significant). Nevertheless, if we observe Figure 2, we can notice how,
also at a glance, there are peaks in tweets trend that precede peaks in price, sug-
gesting a relationship between the two time series. A patent peak of tweets on
11 February, is followed by a growth of Bitcoin’s price. The same circumstance
is visible in the following days: January 23, February 3, February 25 and so on.
We also analyzed tweets with positive mood and we noticed a two-fold increase
in cross-correlation value. Figure 3 shows this result and it’s well rendered that
positive tweets can predict the fluctuations of the Bitcoin’s price. It is proven
by a maximum cross correlation value of -0.35 with a positive delay of almost
4 days. We can confirm that positive mood could predict the Bitcoin’s price
almost 3-4 days in advance. All patent peaks in the positive tweets plot precede
a significant change in the Bitcoin’s price after some days.
Fig. 3. Cross-correlation between positive Tweets and Bitcoin’s price
The cross-correlation result between Google Trends data and Bitcoin’s price
also looks significant. The cross-correlation value increase up to a value of 0.64,
that is quite substantial. This result is shown also by a little significant rela-
tionship that exists between positive tweets and Google Trends data. Figure
4 shows how Google Trends proceeds in the same direction of Bitcoin’s price
and highlighting a striking similarity between them. Table 1 summarizes the
cross-correlation results, obtained comparing the spread among Bitcoin price
and different volumes of data.
Table 1. Cross-correlation results
Compared Systems Cross-correlation value delay
Bitcoin price-Tweets volume 0.15 1
Bitcoin price-Positive tweets -0.35 3-4
Bitcoin price-Google Trends data 0.64 0
Fig. 4. Cross correlation between Google Trends and Bitcoin’s price, expressed in dol-
lars
5 Conclusions
In this paper, we studied whether social media activity or information extracted
by web search media could be helpful and used by investment professionals in
Bitcoins. Since the use of Bitcoins is increasingly widespread, we decided to
analyze the market, in order to predict the evolution of its price.
To this purpose, we presented an analysis of a corpus of tweets about Bitcoin,
considering a total amount of 1,924,891 tweets. The corpus covers a period of 60
days between January 2015 and March 2015. We applied automated Sentiment
Analysis on these tweets in order to evaluate whether public sentiment could be
used to predict Bitcoin’s market. We also used Google Trends media to analyze
Bitcoin’s popularity under the perspective of Web search. In this preliminary
study, we examined the Bitcoin price’s behavior comparing its variations with
these of tweets volume, tweets with positive mood volume and Google Trends
data. From results of a cross correlation analysis between these time series,
we can affirm that positive tweets may contribute to predict the movement of
Bitcoin’s price in a few days. Google Trends could be seen as a kind of predictor,
because of its high cross correlation value with a zero lag. Our results confirm
those found in the previous works, based on a different corpus of tweets and
referred to a different Bitcoin market trend.
While the current data is only 60 days already looks promising, a consecutive
analysis of more than 6 months might provide a better result quality. In further
studies, we also plan to take into account the number of retweets and favorites
for the tweet’s corpus analyzed. Along these lines, we could check whether results
stay unchanged with the addition of this variable.
Acknowledgments. This research is supported by Regione Autonoma della
Sardegna (RAS), Regional Law No. 7-2007, project CRP-17938 LEAN 2.0.
References
1. Kaminski, Jermain, and Peter Gloor. "Nowcasting the Bitcoin Market with Twitter
Signals." arXiv preprint arXiv:1406.7577 (2014).
2. Rao, Tushar, and Saket Srivastava. "Analyzing stock market movements using twit-
ter sentiment analysis." Proceedings of the 2012 International Conference on Ad-
vances in Social Networks Analysis and Mining (ASONAM 2012). IEEE Computer
Society, 2012.
3. Garcia D, Tessone CJ, Mavrodiev P, Perony N. 2014 The digital traces of bubbles:
feedback cycles between socio-economic signals in the Bitcoin economy. J. R. Soc.
Interface 11: 20140623. http://dx.doi.org/10.1098/rsif.2014.0623
4. Mai, Feng and Bai, Qing and Shan, Zhe and Wang, Xin (Shane) and Chiang, Roger
H.L., From Bitcoin to Big Coin: The Impacts of Social Media on Bitcoin Performance
(January 6, 2015). Available at SSRN: http://ssrn.com/abstract=2545957
5. A. Mittal, A. Goel, “Stock Prediction Using Twitter Sentiment Analysis” in pro-
ceeding of IEEE/WIC/ACM International Conference on Web Intelligence and In-
telligent Agent Technology, 2013
6. Bollen, Johan, Huina Mao, and Xiaojun Zeng. "Twitter mood predicts the stock
market." Journal of Computational Science 2.1 (2011): 1-8.
7. K. Dessai and M. Kamat, Application of social media for tracking knowledge in
agile software projects, Available at SSRN 2018845, 2012.
8. Google Trends. http://www.google.it/trends/ . Accessed November 6,2014.
9. Thelwall, M., Buckley, K., Paltoglou, G. : Sentiment in Twitter events. Journal
of the American Society for Information Science and Technology, 62(2), 406-418
(2011).
10. Twitter [online] https://twitter.com/ .Accessed November 6,2014.
11. SentiStrenght. [Online] http://sentistrength.wlv.ac.uk/ .Accessed November
6,2014.
12. Alexa. Top sites: The top 500 sites on the Web. Accessed September 30, 2014, from
http://www.alexa.com/topsites/global
13. Thelwall, M., Buckley, K., Paltoglou, G. Cai, D., Kappas, A. : Sentiment strength
detection in short informal text. Journal of the American Society for Information
Science and Technology, 61(12), 2544–2558 (2010).
14. Java, A., Song, X., Finin, T., Tseng, B. Why we twitter: Understanding microblog-
ging usage and communities. Proceedings of the Ninth WebKDD and 1st SNA-KDD
2007 Workshop on Web Mining and Social Network Analysis (pp. 56–65). New York:
ACM Press (2007).
15. Pang, B., Lee, L.: Opinion mining and sentiment analysis. Foundations and Trends
in Information Retrieval, 1(1–2), 1–135. (2008).
16. Twitter API. https://dev.twitter.com/rest/tools/console . Accessed Novem-
ber 25, 2014.
17. SOCIAL MEDIA AND MARKETS: The New Frontier
18. S. Ye and F. Wu. Estimating the size of online social networks. In Proc. of the
IEEE 2nd Intl. Conf. on Social Computing (SocialCom), pages 169–176, Aug. 2010
19. A. Go, R. Bhayani, and L. Huang. Twitter sentiment classification using distant
supervision. CS224N Project Report, Stanford, 2009.
20. Grinberg, Reuben, Bitcoin: An Innovative Alternative Digital Currency, Hastings
Science & Technology Law Journal, Vol. 4, p.160, Dec 2011
21. Dorit. Ron and Adi Shamir. Quantitative analysis of the full bitcoin transaction
graph.
22. Satoshi Nakamoto. Bitcoin: A peer-to-peer electronic cash system.
23. F. Reid and M. Harrigan. An analysis of anonymity in the bitcoin system.In Proc.
of the Conference on Social Computing (socialcom), 2011.Reid F.Harrigan M.
24. Pak, A and Paroubek, P. (2010) Twitter as a Corpus for Sentiment Analysis
and Opinion Mining. (European Language Resources Association (ELRA), Valletta,
Malta).