Reading Time: 5 minutes

Introduction: why sentiment belongs to election forensics

Election forensics is often associated with numbers: turnout, vote shares, invalid ballots, polling-station anomalies, unusual statistical patterns or inconsistencies between official databases and polling-station records. These are essential elements of electoral verification. However, election results are not produced only on election night. They are produced in a broader political and media environment in which voters form preferences, fears, loyalties and perceptions of legitimacy.

This is why media analysis belongs to election forensics in a broader sense. It does not prove electoral fraud. It does not replace election observation, legal complaints, court procedures or polling-station verification. But it helps identify whether the political terrain has been systematically shaped before voting begins.

The sentiment analysis presented here is based on Serbian terms used in two word clouds: one describing the opposition in pro-government media, and the other describing President Aleksandar Vučić in the same media environment. The contrast is striking. The opposition is described through overwhelmingly negative terms such as “traitors,” “mercenaries,” “terrorists,” “fascists,” “scum,” “enemies” and “monsters.” Vučić is described through overwhelmingly positive terms such as “leader,” “protector,” “winner,” “statesman,” “visionary,” “hero,” “invincible” and “father.”

This contrast matters. Democratic elections require more than accurate counting. They also require a political environment in which citizens can receive pluralistic information and evaluate political alternatives without systematic demonisation of one side and idealisation of another. Sentiment analysis helps make that asymmetry visible.

Sentiment analysis: what it measures and how it can be used

Sentiment analysis is a set of methods used to assess the emotional tone of language. At the simplest level, words, sentences or documents are classified as positive, negative or neutral. More advanced approaches measure intensity, detect emotions such as fear, anger, contempt, trust or admiration, and track how sentiment changes across time, media outlets, political actors and campaign events.

In media monitoring, sentiment analysis can answer important questions. Are political actors described in systematically different emotional terms? Is negative language concentrated around the opposition, while positive language is concentrated around incumbents? Does hostile language increase before protests, crises or elections? Do similar labels appear across multiple media outlets at the same time? Does criticism remain political, or does it shift into delegitimisation and dehumanisation?

Several methodological approaches can be used. A basic approach is dictionary-based sentiment analysis, where terms are coded by polarity and intensity. A more advanced approach uses supervised machine-learning classifiers trained on manually coded media texts. Topic modelling can identify whether negative sentiment is connected to specific themes, such as national security, corruption, foreign influence or public disorder. Network analysis can examine which words co-occur with which actors. Time-series models can track sentiment before and after political events. Multilevel models can separate outlet-level effects from actor-level and event-level effects. Transformer-based language models can classify sentiment and framing at the sentence or paragraph level, provided that they are adapted to Serbian and carefully validated.

For election forensics, sentiment analysis is most useful when combined with other evidence: media exposure, airtime, tone, headlines, source selection, campaign events, public-resource use, voter pressure and election results. It should not be treated as a standalone verdict, but as part of a broader diagnostic toolkit.

The opposition in sentiment analysis: hostility, moral condemnation and threat framing

Table 1: Sentiment analysis of Serbian terms used to describe the opposition in pro-government media
Note: The table presents Serbian terms, English meanings, sentiment polarity and sentiment intensity.

Term	English meaning	Sentiment	Intensity
izdajnici	traitors	Negative	Very High
plaćenici	mercenaries / paid agents	Negative	Very High
teroristi	terrorists	Negative	Very High
fašisti	fascists	Negative	Very High
ekstremisti	extremists	Negative	High
huligani	hooligans	Negative	High
ološi	scum / riffraff	Negative	Very High
bitange	thugs / rascals	Negative	High
lažovi	liars	Negative	High
neprijatelji	enemies	Negative	Very High
idioti	idiots	Negative	High
blokaderi	blockaders	Negative	Medium-High
đilasovci	Đilasists	Negative	Medium (pejorative)
ludaci	lunatics	Negative	High
špijuni	spies	Negative	High
mrzitelji	haters	Negative	High
monstrumi	monsters	Negative	Very High
kriminalci	criminals	Negative	High
batinaši	goons / club-wielders	Negative	High
prodani	sellouts	Negative	High
anarhisti	anarchists	Negative	Medium-High
besramni / zlo	shameless / evil	Negative	High
lopovi	thieves	Negative	High

The sentiment table for terms describing the opposition shows an overwhelmingly negative pattern. Almost all terms are negative, and many carry high or very high intensity. This is not merely critical vocabulary. It is a language of moral condemnation, social exclusion and security framing.

Terms such as “traitors,” “terrorists,” “enemies,” “spies” and “mercenaries” do not simply suggest that the opposition has poor policies. They place opposition actors in a frame of threat. Terms such as “scum,” “monsters,” “idiots” and “lunatics” do not challenge arguments; they attack dignity and legitimacy. Terms such as “criminals,” “goons” and “thieves” connect political competition with crime and violence.

This distinction is central to electoral integrity. Democratic politics allows sharp criticism. Governments and oppositions may strongly disagree, accuse each other of incompetence, expose failures and challenge policies. But when political opponents are repeatedly described as traitors, terrorists or enemies, criticism shifts into delegitimation. The message is no longer “their policy is wrong.” The message becomes “they do not belong in legitimate politics.”

Such language can affect voters in several ways. It may discourage citizens from openly supporting opposition actors. It may stigmatise civic participation, protest or election observation. It may make institutional inaction appear more acceptable when opposition actors complain. It may also frame demands for verification of election results as politically suspicious before they are even examined.

From the perspective of election forensics, this is a relevant political signal. It does not prove manipulation of election-day procedures. But it suggests that the pre-election information environment may be systematically unequal. If one political side is repeatedly pushed into categories of betrayal, criminality and danger, then its formal right to compete remains, but its practical ability to compete on equal terms is weakened.

Vučić in sentiment analysis: positive idealisation and personalised leadership

Table 2: Sentiment analysis of Serbian terms used to describe President Aleksandar Vučić in pro-government media
Note: The table presents Serbian terms, English meanings, sentiment polarity and sentiment intensity.

Serbian term	English meaning	Sentiment	Intensity
lider / vođa	leader / chief	Positive	Very High
zaštitnik	protector	Positive	Very High
snažan / jak	strong	Positive	High
pobednik	winner	Positive	High
uspešan	successful	Positive	High
patriot	patriot	Positive	High
mudri	wise	Positive	High
hrabri	brave	Positive	High
vizionar	visionary	Positive	High
državnik	statesman	Positive	High
odlučan / nepokolebljiv	decisive / unwavering	Positive	High
neuništivi / nepobedivi	indestructible / invincible	Positive	Very High
čuvar	guardian	Positive	High
reformator	reformer	Positive	Medium-High
heroj	hero	Positive	High
najbolji	the best	Positive	High
posvećen / verni	dedicated / loyal	Positive	High
radnik	hard worker	Positive	Medium-High
inteligentan	intelligent	Positive	Medium-High
skroman	humble	Positive	Medium
beskompromisni	uncompromising	Positive	Medium-High
istorijski / genijalni	historic / genius	Positive	High
otac	father (of the nation)	Positive	High
neprikosnoveni	untouchable	Positive	Very High
predsednik	president (elevated)	Positive	High

The second sentiment table shows the opposite pattern. Terms used to describe President Vučić are overwhelmingly positive, often with high or very high intensity. They construct an image of strong, successful, protective and historically significant leadership.

Positive coverage of a political leader is not, by itself, a problem. Media outlets may praise a politician, support a government or highlight achievements. The electoral-integrity concern arises when this strongly positive framing of one actor is combined with systematic negative and dehumanising framing of his opponents. In that case, the public sphere does not function as a space for comparing programmes, policies and records. It becomes a moral binary: protector versus enemies, stability versus chaos, leader versus traitors.

Terms such as “leader,” “chief,” “protector,” “statesman” and “guardian” frame Vučić as a source of security. Terms such as “winner,” “successful,” “invincible” and “indestructible” frame him as politically inevitable. Terms such as “hero,” “visionary,” “genius” and “father” move beyond ordinary political approval into a highly personalised symbolic register.

For voters, such language can influence the emotional environment of choice. Voters do not make decisions only by comparing manifestos or economic indicators. They also respond to trust, fear, loyalty, perceived competence and perceived risk. If one actor is repeatedly presented as the only guarantor of stability, while alternatives are described as dangerous or illegitimate, the electoral choice becomes emotionally structured before the campaign formally begins.

This is why sentiment analysis is relevant to election forensics. It shows the emotional architecture of political communication. It helps identify whether citizens are exposed to a balanced contest among legitimate alternatives, or to an information environment that systematically elevates one side and delegitimises the other.

Conclusion: sentiment as evidence of an uneven political terrain

The sentiment analysis of the terms used in the two word clouds reveals a strong asymmetry. Opposition actors are described through negative, often highly intense labels associated with betrayal, crime, violence, foreign influence and hostility to the state. President Vučić is described through positive, idealising terms associated with strength, success, protection, wisdom and national leadership.

This asymmetry matters for elections. Voters do not decide in an empty space. They decide within an information environment that defines what is normal, what is dangerous, who is trustworthy, who is threatening and who is legitimate. If that environment consistently presents one side as a protector and the other as a threat, the political terrain becomes uneven before the first polling station opens.

Sentiment analysis does not determine how individual voters will behave. It does not prove that votes were manipulated. But it does show how the language environment may shape voter perceptions, social pressure and perceived legitimacy. It helps explain why election forensics cannot be limited to election-night arithmetic.

A verifiable election requires more than accurate polling-station records. It also requires scrutiny of the conditions under which voter preferences were formed. Sentiment analysis offers one way to examine those conditions: by measuring the emotional imbalance in media language and linking it to the broader question of democratic integrity.

If elections are to be genuinely verifiable, we need to check not only the count, but also the terrain on which political choice was formed.