English. Relying on linguistic cues obtained by means of structural topic modelling as well as descriptive lexical analyses, this study contributes to the general understanding of the Twitter users' response to the annual Italian budget law approved at the end of December 2018. Some topics contained in the dataset of tweets are procedural or generic, but besides those, it often emerges that Twitter users expressed their concern with respect to the provisions of this law. Supportive attitudes seem to be less frequent. This paper also advocates that findings from inductive studies on Twitter data should be interpreted with caution, since the nature of tweets might not be adequate for drawing far-reaching generalisations.
In the last decade, Internet has revolutionized human communication and interaction. And among all forms of digitally-mediated communication, social media stand out as one of the most effective. As Boulianne (2017) points out, the effects of social media depend on their nature of use (e.g. source of information; one-to-one/one-tomany/many-to-many communication; networking and relationship-building; expression of opinions; etc.).
Nowadays, potentially everyone with a computer or a mobile device having access to the internet can write and share contents which may be viewed and debated immediately by other people. The impact of a social media post may be huge, and unlike other prior forms of communication, it can easily cross borders in just a few seconds. In fact, social media make things happen faster than ever before. For instance, Facebook and Twitter were crucial in allowing the Arab uprisings or the Romanian anti-corruption protests to happen more efficiently and on a larger scale. 2
Besides their essential role in information
dissemination, networking, and people mobilization,
social media are also important indicators and
predictors of their users’ opinions, sentiments and
attitudes. In fact, various studies have explored
people’s reactions towards social, economic, and
political issues, by analysing social media posts
With over 6,000 tweets posted every second,
corresponding to roughly 350,000 per minute, 500
million per day, and around 200 billion per year,
Twitter has become one of the main tools of
communication worldwide
This means that Twitter may represent an easily
exploitable opportunity for politicians in their
attempt to reinforce communication with potential
voters in what might be defined as a permanent
digitally-mediated electoral campaign.
Additionally, it has been suggested that Twitter could be
used to model and predict public opinion and
behaviour regarding political events, such as
electoral campaigns
While indeed tweets have been widely used to
analyse public opinion and political discussions in
all its forms, several methodological
considerations are dutiful. First of all, Twitter users do not
represent an optimal sample for public opinion or
voting population, especially due to their higher
than average level of education and political
sophistication, as well as a generally younger age
Acknowledging all the limitations mentioned
above, this inductive exploratory study aims to
contribute to the growing body of literature
examining Twitter and its increasingly prominent role
in online communication by studying its
application in the context of political discourse. In
particular, the linguistic approach presented here is
providing insights into tweets regarding the
discussion and the approval of the annual Italian
budget law (in Italian “legge finanziaria” and/or
“legge di bilancio”). This law was also often
labelled as “the manoeuvre” (in Italian “la
manovra”) and “the people’s manoeuvre” (in
Italian “la manovra del popolo”) by its proponents –
in particular Movimento 5 Stelle (abbreviated
M5S) –, mainly due to some of its populist
provisions (e.g. the citizen's basic income and pension).
3 The full text of the annual Italian budget law
This topic is worth studying since the two parties displayed differences in economic, fiscal, infrastructural, and social policies both in the electoral campaign for the 2018 general elections as well as during the first months of government. For instance, Lega supported the flat taxation on incomes, while M5S the citizen's basic income (“reddito di cittadinanza” in Italian). However, these measures, although slightly modified, as well as the amendment to the 2011 pensions reform (“quota 100” in Italian) were included in the coalition agreement and subsequently in the draft for the annual budget law. The bill also contained various other economic and fiscal provisions (e.g. taxes on digital services; new VAT rates; reducing military expenses and the Italian contribution to United Nations; new labour measures; environmental incentives; etc.)3.
We believe that the textual material contained in tweets may be promising in providing hints on how Twitter users – a fraction of the Italian voters – reacted to the provisions of the budget law. Linguistic insights into tweets might be able to guide us in understanding whether the so-called “manovra del popolo” was perceived by Twitter user as representing indeed the people’s interest. 4
Although in the Western world there are three
mainstream social media networks (i.e. Facebook,
Instagram, and Twitter), in this paper we analyse
Twitter posts, primarily as a consequence of data
availability. Indeed, unlike other tools for social
media, Twitter APIs for R
Using the rtweet package
To avoid duplicates, we discarded all retweets and all posts that contained quotes of other tweets. The removal process was obtained by filtering the dataset, thus selecting only tweets whose values for “is_retweet” and “is_quote” corresponded to “FALSE”. Duplicates other than retweets and quotes were removed with R’s base functions duplicated – which identified duplicated tweets – and unique – which extracted unique tweets. Since the aim of this study is to uncover the reactions of the Italian voters active on Twitter, we removed the tweets written by political actors. To do so, we defined a list containing the Twitter usernames of the members of the Italian Parliament, as well as those of the official national and local party profiles; this list was used to automatically filter and remove tweets published by the unwanted profiles. We decided to keep tweets from news agencies, online newspapers, and television channels, since they could represent vectors of information exchange regarding the topic analysed in this study. The final dataset contained 20,891 tweets.
Tokens 701,986
414,803 75,485
Lemmas 31,947 Table 1: Dataset statistics. 4.1
Since the tweets and their metadata would have
been used for lexical analyses and structural topic
modelling5, we performed several pre-processing
steps: defining a “stop words” list for Italian
consisting of roughly 1,000 lexically empty or
uninformative words (i.e. prepositions, conjunctions,
auxiliary verbs, etc.); uniformizing, normalising
and cleaning the texts with various corpus
processing functions available on the R packages
quanteda
As a result of the ever-growing interest and
availability of text data – often unstructured –, various
statistical and machine-assisted approaches for
the analysis of textual material have been
proposed. In this paper we are employing the
Structural Topic Model (STM) – a generative model of
word counts –
Similarly to Latent Dirichlet Allocation
After having employed the STM’s searchK
function to perform several tests, such as held-out
likelihood and residual analysis, the ideal number of
topics seemed to be between 10 and 14.
Additionally, STM gave the possibility to set the type of
initialization, so here the spectral one was chosen,
since previous studies had proven its stability and
consistence
Figure 1 in Appendix shows the topics related
to the annual Italian budget law as they emerged
from the analysis of tweets. Each topic was further
classified into one category (i.e. EU &
Confidence, Main Measures, Criticism & Concern,
Government vs. Opposition, Procedures –
Generic, Support). This classification was based on
the correlations obtained from a hierarchical
clustering representation performed with the plot
function of the stmCorrViz package
Although we do not claim to model public
opinion from tweets, interestingly, the topics
managed to echo various issues regarding the
budget law. Judging by the expected topic
proportions, one could order the most prevalent topics as
follows: Topics 9, 8, and 3 (sum of topic
proportions: 0.29) reflect disapproval and doubts
towards the provisions of the budget law; Topics 1
and 7 (sum of topic proportions: 0.22) describe the
difficult negotiation with the European Union
(EU) and the threat of an infringement procedure;
Topics 10 and 2 (sum of topic proportions: 0.19)
depict the main measures contained in the budget
bill; Topic 6 (topic proportion: 0.13) illustrates the
support to the budget bill and to the Government;
Topic 5 (topic proportion: 0.11) refers to the
procedures regarding the discussion, the vote, and the
approval of the budget law; and Topic 4 (topic
proportion: 0.06) reveals the conflict between the
content represents the words used to discuss about that
topic
Government and the oppositions on the provisions of the law.
After having calculated the estimated effects of the temporal covariate on topical prevalence, a plot displaying this variation was created. Figure 2 in Appendix shows how the afore-mentioned topics varied over the 43 days considered. Topics are ordered as a function of their expected proportions.
Firstly, there emerged that the variation was not particularly strong, except for some topics. For instance, Topic 9 had a peak at the end of December/the beginning of January, suggesting that Twitter users might have written tweets of concern soon after the approval of the annual Italian budget law. On the other hand, Topic 6, which contained mostly tweets of support towards the measures of the budget bill seemed to be prevalent primarily at the end of November and in mid-December. The procedural topic was generally prevalent at the end of December, a timeframe corresponding to the vote and approval of the law. The two topics summarising the negotiations with the EU, the confidence, and the possible infringement procedure were pervasive during the entire period considered, with some peaks in early- and midDecember. Topic 4 that regarded the disagreement between the Government and the opposition was constant over time, and so were the topics delineating the main measures of the law. 5.2
We were also interested in performing descriptive
lexical analyses on tweets. First of all, with the
terminology extraction tool on Sketch Engine
Syntagma reddito di cittadi
nanza procedura di infra
zione clausole di salvaguardia
the citizen’s basic
income infringement pro
cedure safeguard clauses voto di fiducia blocco assunzioni professioni sanitarie senza titolo
flat tax commissione bilan
cio gilet azzurri taglio pensioni scatoletta di tonno governi precedenti pensioni minime scatola chiusa
nuove tasse promesse elettorali
fasce deboli deficit strutturale accordo tecnico braccio di ferro appalti senza gara
assurdità totale terrorismo media
tico auto inquinanti più tasse confidence vote
hiring freeze health professions without a degree
flat tax budget committee
blue vests pension cuts
tuna can previous govern
ments minimum pensions black box new taxes campaign promises vulnerable citizens structural deficit technical arrange
ment trial of strength no-bid contracts total nonsense media terrorism polluting cars
more taxes sovereignist govgoverno sovranista ernment manovra contro il manoeuvre against popolo the people false promesse false promises IVA sui tartufi VAT for truffles popolo italiano Italian people Table 2: The most representative syntagmas in the dataset.
It is clear that various multi-word expressions referred to procedural aspects, such as those reflecting the vote and the approval of the budget law (e.g. “confidence vote”), while others were used to list its measures, especially fiscal and economic policies (e.g. “the citizen’s basic income”, “flat tax”, etc.). Nevertheless, various syntagmas seemed to express doubts with respect to the provisions of this law. In fact, often, the words chosen by many Twitter users to express their criticism were rather strong (e.g. “total nonsense”, “black box”, “sovereignist government”, etc.).
These concerns and rather negative reactions to the budget bill were reflected also in the word sketches (i.e. visual representations of collocations and word combinations obtained on Sketch Engine) for the words “manovra” and “legge”. Generally, three different scenarios are distinguishable.
First of all, there were several neutral verbs, nouns, and modifiers associated to the budget law, most of which regarding its procedural aspects. The most frequent (i.e. frequency ≥ 10.81 per million) are listed below:
Word/Syntagma English scrivere write cambiare change modificare modify discutere discuss approvare approve contenere contain prevedere consist varare launch votare vote passare pass riscrivere rewrite promulgare promulgate gialloverde yellow-green economica economic finanziaria financial populista populist discussione discussion commissione commission bilancio budget
Table 3: Neutral associations.
Next, some positive evaluations of the budget law emerged. The most frequent (i.e. frequency ≥ 10.81 per million) are listed below:
Word/Syntagma TraEnnsglalitsiohn into favorire (l’innova- favour (innovazione) tion) grande big buona good bella beautiful significativa significant del popolo of the people del cambiamento of the change per i cittadini for the citizens per la crescita for the growth
Table 4: Positive associations.
Nonetheless, several word associations seemed to suggest negative reactions to the budget law. The most frequent (i.e. frequency ≥ 10.81 per million) are shown below:
recessiva piena di errori
recessive full of errors This paper explored the Twitter users’ reactions to the annual Italian budget bill. STM outputs and descriptive lexical analyses showed that tweets concerned various aspects associated to the object of this study. Apart from talking about procedural and generic issues, users expressed their doubts and disapproval with respect to the measures of the budget law. Generally, tweets supporting this law were less frequent. The findings of this study, although preliminary, might be seen as indicators of what subsequently turned out to be a failure for the first Conte government. Still, as reiterated throughout the paper, the results might not reflect the real attitudes of the Italian voting population, since Twitter users tend to be younger and to have an above the average level of education and political sophistication (cf. Barberá et al., 2015). Moreover, tweets, by nature, might not be suitable for drawing steady generalizations, even if the prospects they offer for content and discourse analysis are indeed significant. Further research on this topic might include the investigation of Twitter user’s reactions by means of sentiment analysis.