The integration between o cial statistics and social media data is a challenging topic. This contribution aims to present a recentlydesigned framework to compare sentiment analysis on social media content with social and economic data. Such framework - which has already been applied, in a preliminary fashion, to the Felicitta project - is meant to integrate o cial statistics and correlate it with online social media data. Its ultimate goal, in fact, namely consists in giving a contribution to the de nition of a measure of subjective well-being that could fully bene t from both traditional, well-established social indicators and dynamic data obtained from the web.
The signi cant growth of user-generated content on the web, and in particular the increased availability of data from online social media, has fostered the development of automatic techniques for the extraction and processing of such content for di erent purposes. This development is re ected, among other things, by the spread of scienti c contests whose main track is the Sentiment Analysis (henceforth SA) of texts in di erent languages (see eg. SemEval1 for English, and SENTIPOLC@EVALITA20142 for Italian).
In turn, these achievements are encompassed into a broader and interdisciplinary debate related to the study and de nition of measures that could be considered as reliable indicators of the well-being of a community. In fact, a growing debate has recently involved the measurement of social and individual well-being. New statistical measures have been proposed besides the bare Gross Domestic Product (GDP), traditionally seen as the best way to measure national
2 http://www.di.unito.it/~tutreeb/sentipolc-evalita14 economic results. Among such measures are a large amount of indicators that, in several ways and from di erent points of view, attempt to assess the degree of \happiness" and life satisfaction, also designated with the expression Subjective Well-Being, or simply SWB (see Section 2). Such measures are usually provided by governmental institutions or entitled research organizations, and they generally include social indicators measuring life quality and concerning all major areas of citizens' lives. However, such data are static and their recovery may require much e orts in terms of time and resources. Moreover, the increasing success of Sentiment Analysis (SA) techniques on social media has made it possible to develop alternative tools and measures, with respect to the latter, to assess the degree of happiness and well-being. Social media and their content can thus be used to complement and corroborate the information gathered from traditional data sources as regards SWB detection.
The work presented here is just part of this research context. In particular,
the purpose of this paper is to describe a framework whose entire de nition and
completion is still in progress, for the analysis and assessment of the degree of
\happiness" of a given community in Italy, taking into account and combining
together the information gathered from two main data sources: a) social media
content, and Twitter in particular; b) the socio-demographic information made
available by the main suppliers of o cial statistical data, such as the Italian
National Institute of Statistics (ISTAT)3. The rst point in particular has
actually been explored and developed under a recent project called Felicitta4 [
The present work is both an extension and a comprehensive reference framework of that project. As a matter of fact, its aim is manifold and includes: 1) the use and further development of techniques for the visualization of SA outcomes in Italian texts; 2) the study of the correlations between o cial statistics and user-generated media content; 3) providing a contribution to the debate on what can be considered e ective and reliable indicators of social well-being.
The remainder of the paper is structured as follows: Section 2 provides a brief introduction to the notion of subjective well-being, summing up the more recent work carried out on this matter while Section 3 describes the whole architecture of the system, as currently conceived. Final remarks in Section 4 close the paper. 2
The present contribution covers the debate on the Subjective Well-Being as a social indicator and sheds some light on happiness studies based on the sentiment analysis of social media.
Subjective Well-Being. As stated in Diener [
The well-being of the population: from Easterlin to GNH. The Gross
Domestic Product (GDP) is today the main measure of the nation's economic activity.
However, since late 70's, a huge debate has grown over this measure [
Social Media and Well-Being. The analysis of textual expressions in social
media contents on a Big-Data scale would o ers an opportunity to economists
and sociologists in the measurement of social well-being. There's an open debate
on the topic and several works already investigated this subject with contrasting
results. Wang et al. [20] examine Facebook's Gross National Happiness (FGNH)
indexes and Diener's Satisfaction with Life Scale (SWLS), and nally criticize
the idea that a well-being index can be based on the contents of a speci c online
social networks. Quercia et al.[17] explore the relationship between sentiment
expressed in Twitter messages and community socio-economic well-being and,
on the contrary, they found interesting correlations between sentiment and
general well-being. Kramer [
The challenge of visualization. The lecture of patterns and trends from
spreadsheets or lists of numbers is a di cult task when we have to deal with large
amounts of data. An improvement is often obtained by the use of graphs. Shapes
and lines immediately create meanings and signi cance from data. In this way,
data visualization allows us to present trends, to discover what is often hidden [
The present framework aims to include several approaches and techniques in order to detect the well-being of a community under a broader perspective. The steps entailed in the design phase was: a) the de nition of the whole pipeline; b) the selection of data from o cial statistics to be correlated with the analysis performed by the SA module; c) the presentation of the most promising patterns emerging from the comparison between social media data and o cial statistics. In this section, we describe the general framework architecture with an overview of its modules. 3.1
The whole framework architecture, as shown in Figure 1, consists of 5 main parts: Providers, Data Gathering, Data Analysis, Data Exposure and Data Visualization.
Providers. Providers are the data sources: i.e Twitter, from which we retrieve
the geolocated8 Italian tweets using the Stream API, and the various
sociodemographic data sources (detailed in Table 1), that return demographic and
socio-economic variables of di erent Italian administrative divisions.
Data gathering. This module is further divided in submodules, each one tackling
one particular task:
{ the Collect submodule collects data from di erent providers;
7 http://twitter.github.io/interactive/sotu2015/
8 For details on the geolocalization methods used, see [
Data Analysis. First, the Sentiment Analysis submodule returns for each
tweet a mood value (positive, negative, neutral); the SA engine is the one
developed in Felicitta, as described in [
Data Exposure. A web server exposes elaborated data by REST API. When a client runs a query, the server queries the database and returns the response.
Data Visualization. Finally, a web client presents the data obtained as response to the queries. For the time being, the visualization module allows to browse either the sentiment data (as in the example in Figure 2), or the sentiment data combined with demographic data, as shown in Figure 3. The part that shows socio-demographic statistics and correlations is yet to be completed. 3.2
As a measure of the mood related to an area in a given period, we consider the percentage of positive tweets. In order to relate moods and numeric social indicators in di erent administration degrees, in Table 1 we summarized some social indicators that could provide an overview of the social well-being of a given community.
As data collection is not always an easy task and the Open Data is not yet widespread in Italian public administration, we realistically decided to focus our attention on data that could be easily accessed and retrieved from public administration web-sites. In order to detail di erent aspects of the society, we resort to di erent sources. In this way, we consider data from di erent elds and viewpoints, mainly demographic and economic.
As regards the demographic eld, the main aspects considered are nationality, gender, age and marital status, since they are closely related to the perception of social well-being. We are interested, for example, in understanding whether and to what extent nationality may in uence the sentiment expressed through social media, or whether married men are happier than singles.
Concerning the economic eld, we consider both jobs data (e.g. the unemployement rate) and data about companies (e.g. enterprises demography). Our hypothesis is that people express negative sentiments more likely if they live in an area with signi cant unemployment rate or with a greater amount of cessations of business.
We also collected data about the real estate market, that we consider a typical indicator of the wealth of a territory. A correlation, in fact, is expected between this aspect and the overall mood detected in social media: the higher the prices (then the wealthier the area considered) and the greater the happiness may be. Similarly, we consider the amount of deposit and loan from the Bank of Italy as a measure of both individuals and public wealth. We selected this set of data as they are representative of di erent relevant social needs, with di erent time and granularity. A rst integration between social media data and demographic data is shown in Figure 3.
Our current work then namely consists in exploring all the possible correlations between the indicators mentioned above and the output of the SA engine, and in improving the visualization module so as to better highlight such correlations and emerging patterns. 4
In this paper we introduced an ongoing project on a framework for the analysis
and assessment of the degree of \happiness" of a given Italian community,
taking into account data from o cial statistics and SA data obtained from social
media. We noticed at least two main problems: the representativeness of data
and the role of ironic sentences. First, the di usion of internet and the use of
online social networks is not widespread in the same way over all kinds of
population. Therefore, for instance, the sentiment of poorest people and elderly can
be not represented or largely underrepresented. This is a classical problem of
quite every sociological inquiries, mainly solved by representative sampling and
qualitative research. A second issue is the presence of irony where the unintended
meaning of words can often reverse the polarity of the message. We well know
this problem and we state how exists a growing interest in this research subject,
as we already investigate the role and the detection of irony and sarcasm[
Acknowledgments. Part of the present contribution has been awarded in the 2014 Istat-Google Contest on \Producing o cial statistics with Big Data": http: //www.istat.it/it/archivio/144042 12. A. Kramer. The spread of emotion via facebook. In Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems, pages 767{770. ACM, 2012. 13. E.W.Dunn L.B.Aknin, M.Norton. From wealth to well-being? money matters, but less than people think. The Journal of positive psychology, 4(6):523{527, 2009. 14. L. Mitchell, M. Frank, K. Harris, P. Dodds, and M. Danforth. The geography of happiness: Connecting twitter sentiment and expression, demographics, and objective characteristics of place. PLoS ONE, 2013. 15. T. Nguyen, B. Dao, D. Phung, S. Venkatesh, and M. Berk. Online social capital: Mood, topical and psycholinguistic analysis. Proceedings of the Seventh International AAAI Conference on Weblogs and Social Media, 2013. 16. OECD. OECD Guidelines on Measuring Subjective Well-being. OECD Publishing,
Paris, 2013. 17. D. Quercia, J. Ellis, L. Capra, and J. Crowcroft. Tracking gross community happiness from tweets. Proceedings of the ACM 2012 conference on Computer Supported Cooperative Work (CSCW '12), 2012. 18. Ben Shneiderman. The eyes have it: A task by data type taxonomy for information visualizations. In Visual Languages, 1996. Proceedings., IEEE Symposium on, pages 336{343. IEEE, 1996. 19. C.o Wang, Z. Xiao, Y. Liu, Y. Xu, A. Zhou, and K. Zhang. Sentiview: Sentiment analysis and visualization for internet popular topics. Human-Machine Systems, IEEE Transactions on, 43(6):620{630, 2013. 20. Kosinski M. Stillwell D.J. Wang, N. and J. Rust. Can well-being be measured using facebook status updates? validation of facebook's gross national happiness index. Social indicators research, pages 483{491, 2014. 21. N. Yau. Visualize this: the FlowingData guide to design, visualization, and statistics. Wiley Publishing, 2011.