Big Data refers to an enormous volume of structured and unstructured data that cannot be handled with traditional databases. The emergence of big data is due to the huge quantities of information. Currently, researchers tend to study big data particularly data produced from social networks. Nowadays, these latest become a vital and crucial tool in tracking and extracting public opinion for developed countries o↵ ering benefits to the democratic process of election. In this paper, we aim to identify the political preferences and tendency of the Tunisian population using classification and opinion mining techniques. To prove the usefulness of the proposed method, we analyze electoral data sets in Tunisia obtained from the o cial sites of independent higher instance for election. This analysis demonstrates close correspondence between election results and extracted opinion.
Big data is an extensively used term in
tremendous data miscellany. Thus, this huge amount of
data makes hard and sometimes impossible the
analysis in a comfortable way using traditional
data processing techniques
The exponential rise of social networks popularity influences real-world politics. These platforms have been exploited in Arab revolution as a mobilization tool and showing social movements. Moreover, social networks are used to study voters preferences in order to forecast electoral results. We cant ignore the role of social networks especially Facebook in our election, using opinion mining we can analyze and predict the voting intention of Facebook users in Tunisia for the 2014 presidential election. The huge amounts of public opinion data need accurate tools to extract useful information. In this study, we exploit stables and some statistical o cial data from ISIE in order to identify possible correlations between socio-economic features of voters and elections results. Furthermore, to well draw voting behavior, we study psychological sociological and economic variables. Thus we have discovered high correlation. The exploited data consists of the results of the Tunisian election of 2014 and 2011, also as we said previously some socio-economic variables and some data from social networks especially Facebook and Twitter. In fact, we treat Facebook activity data during months preceded the legislative and presidential elections. An important part of our study is to determine positive and negative opinions. These opinions can then be interpreted to define the political position of voters. Persons who share several common characteristics present a tendency to have common voting behavior. We are interested in establishing voting behavior according to voter0s profile. The remainder of this paper is as follows: in part 2, we discuss existing works handling Socioeconomic variables a↵ ecting elections results and voter0s sentiment analysis. In part 3, we present data and methodologies use in this paper in order to present opinion mining patterns of voters in the Tunisian election in part 4. Finally, an overview of our work and opportunities that has been opened up are provided in part 5.
Related Works Before presenting the use case of this paper, we flew over existing works handling opinion mining voters patterns. Two sides were manages: • Patterns a↵ ected by socio-economic variables • Voter0s sentiment analysis 2.1
Socio-Economic Variables
The paper develops in
We should firstly define sentiment analysis called
also opinion mining, it is considered as a new
text analysis
Data and Methodology
To reach our objective, we have chosen the
results of the 2014 parliamentary and presidential
elections in Tunisia. The election results were
fairly contested and were not surrounded by any
extraordinary events. The sources of our data are
the web site of the Independent Higher Instance
for Election (ISIE) 2014 and the National
Institute of Statistic (INS). The two references report
the data from 24 governors and 264 delegations
for the legislative, the first and second round in
presidential election. We collected also Twitter
and Facebook information from the start of the
legislative election on October 2011 and the
electoral campaign from August 2014 to the end of
second round of presidential election. Political
tweets and Facebook comments includes those
coming from followers and candidates. To
analyze data collected from ISIE and INS, we
apply WEKA algorithms. A study in
People discus and express their opinion by commenting, posting or sharing status (or tweets). We are interested only to status, tweets and comments containing keywords related to election, parties and candidates manually found. People publications may contains emoticons and informal language, thus we should analyze those manually. After finding appropriate status or comments referring to our subject, the collected data consists of 274 status and 586 comments. • Step2 : Status Classification
Counting the number of positive, negative and neutral messages. Actually, using opinionFinder tool we are able to extract automatically subjective sentences and sentiment expressions. Based on SentiStrength which estimates the strength of positive and negative sentiment in texts and emoticons in various languages, we classify words into categories.
To design the voter profile, our system is based on identifying his sentiment. Consequently, we can conclude if a given voter is with or against party or candidate. Additionally, we can interpret his preferences according to his opinion. The aim of this work is to determine whether social networks data, and the data gathered from the ISIE o cial site can predict election results. The figure below illustrates the proposed methodology 1:
Empirical Results
Voter Behavior The objective of this paper is to extract relevant and interesting knowledge from big amount of data using supervised learning techniques. when applying our chosen methodology, our important finding is that voters distinguish between major and minor parties in a governing coalition and hold only the major incumbent party accountable for economic growth.
There is a tendency for parties to lose or win portion of their votes between first and second round due to political mergers which eventually happen in the second round. This refers to voter turnout patterns which are: • Staying put: a voter remains a liated with his her party in legislative and both rounds of presidential election. • Party Switching: di↵ erent from fusion (party0s voters re-a liates in the form of a merger with another party). • Starting up: una liated voters change their opinion to be a liated to another party. We have found that Islamist party represented by Mohamed Moncef Marzouki (MM receives their votes from people living in areas su↵ ering from lack of economic growth such as Sidi Bouzid, Kebili... NIDAA TOUNESS represented by Beji Caied Sebssi (BCE) received their votes from those living in regions experiencing good economic conditions and economic growth like Tunis, Nabeul 2.
The recent election shows a decrease in NAHDHA (Political Party) partisans instead of an increase in NIDAA TOUNESS(Political Party). This is due to several reasons: • A tendency to vote against the parties holding power: voters who votes BCE to decrease the power to MMM. • Government0s economic performance • Decrease of Quality of life • Increase of poverty rates.
We observed that citizens of a minor socioeconomic status don0t exhibit political e cacy and commonly ”participate less in the voting process”. In contrast, citizens with important socioeconomic status participate and sponsor the political process. Thus, voter0s income and occupation are crucial in vote0s process. We mention that high education level and low level of illiteracy are behind significant civic activity 3. We found also that woman voted more to BCE. 4.2
Tunisian presidential election 2014 in the second round was limited between two main candidates where the concurrence was severe and intense. Actually, most sentiment and opinion have been predicted and discovered according to online discussions, status and comments. Collecting and analyzing sentiment is essential for electoral campaigns. Indeed, the enormous quantity of data related to public sentiment and opinion is considered as big data which played a crucial role for Obama campaign.
With the accessibility of social networking sites,
internet users can easily express their opinions
about various topics. Thus, we can exploit and
mine the shared opinions. In this work, we
collect data from Facebook, we focus on the content
of status containing a reference to political party
or also politician
Positive
Caid Sebssi est une chance pour la Tunisie Des tunisiens craignent un retour en arrire aprs lection de BCE! BCE: Rencontre avec les prsident des clubs sportifs ures below (figure 4 and 5) provide a curve that mentions the evolution of the number of tweets between August 15 and September 14, 2014; we recorded 2,004 tweets containing the hashtag TnElec. This hashtag is sometimes used more than 150 times a day for the last few weeks and also an overview about the number of Facebook users that like the candidates o cial pages. The x axis represent the number of negative and positive statues written for both presidential candidates The number of negative status for MMM is larger comparing to BCE according to the collected data (figure 6) which reflects the reality i.e. ers is a↵ ected by economic, demographic and socio-economic characteristics.
We are aware that some interpretations require additional data. Interpretations presented in this work can serve as hypotheses in next elections. In bulk, despite the limits of social networks analysis, we can assert that our results are encouraging.
De´tection