Personality on Twitter

Psychological traits of the most-followed Italian users

Domenico Bianco, Mauro M. Gentile, Francesco Grisolia | 11 luglio 2017

Tags: personality, Twitter, Big Five, LIWC

Our online behaviour reveals our character. The traces we leave on social networks say a lot about us, showing our interests and outlining our personality traits. The aim of “Italians on Twitter” has been to show how much our tweets reveal about ourselves

We carried out the project during our master in Big Data Analytics and Social Mining at the University of Pisa.

Twittersphere and the Big Five Model

We used two main tools: LIWC, an automated text analysis program, and the Big Five Model. The latter allows to detect psychological profiles through the combination of five basic traits: openness, conscientiousness, extroversion, agreeableness, neuroticisim. Both are well-established tools in personality detection studies.

The detection of personality traits using social networks is a new and expanding research field, in which psychological theories, psychometry, computational linguistics and machine learning techniques converge.

There has been a hot debate on users, consumers and voters profiling over the last years. Cambridge Analytica, the London-based political consulting firm, notoriously employed The Big Five Model during the Brexit and the last US presidential campaigns. However, its crucial ability to micro-target and influence British and American voters is among the most controversial issues.

In order to evaluate the real effectiveness of such approach, we aimed at outlining the psychological profiles of Italian Twitter users through the Big Five Model. We took into account that focusing on Twitter could be a tricky task: tweets are shorter than Facebook posts, therefore they generally offer a smaller amount of data for automated textual analysis. Such aspect is however compensated by Twitter’s higher accessibility through API, compared to Facebook.

We also asked ourselves to what extent social media writing can reflect authentic psycological dispositions. Recent studies, however, underline that our on- and offline activities are more and more intertwined, blurring the line between the two dimensions.

How to analyse 1,7 billion tweets

We used Python for textual analysis and D3js for data visualization. Our database – 14,2 million unique accounts and 1,7 billion tweets – has been created and then analysed between April and June 2017. In order to obtain reliable results we focused on a subset of 2,5 million accounts, whose tweets offered at least 70 matches with the LIWC dictionary.

Having defined our target universe, we first singled out the 1000 most-followed Italian Twitter users. Then we crawled their accounts and extracted their followers and followings, thus expanding our dataset. Finally we used text mining techniques to detect the associated gender and location – the province or region in which a user lives or from which he/she just uses Twitter.

Planners living in the North, extroverts in the South

The geographical distribution of psychological traits among Italian users seems to confirm some common places on our country, as the propensity to plan in the North-east, the general extroversion in the South and the existence of two heavens of organization and cooperative spirit, Aosta Valley and Trentino-Alto Adige.

GEOGRAPHY OF PERSONALITIES ON TWITTER. The acronym OCEAN stands for the 5 psychological traits on which the Big Five Model is based: Openness, Conscientiousness, Extroversion, Agreeableness, Neuroticism.

Affinities and interactions on Twitter

We singled out five clusters among Italian users, that is five groups with distinctive values and combinations of traits. As the visualization below shows, two couples of clusters mirror each other: the disorganized and focused ones, on the one hand; innovators and conservatives, on the other. In the middle we find the balanced ones, characterised by average values in all 5 OCEAN traits.

Their ways of interactions also turned out to be significant. A mirror effect is quite evident: there is a general inclination to follow users with similar psychological profiles. At the sime time, a certain attraction to diversity can be noticed: the balanced users tend to follow, share tweets of, and mainly quote the disorganized ones.

The most-followed have a more defined personality

The results obtained through the analysis of the 4000 most-followed accounts reveal interesting affinities in different categories of users: politicians, communication experts, show business figures, Twitter stars, musicians, sportspeople and the accounts we labelled as “impersonal” – that is companies, public authorities, media outlets (as newspapers or tv channels).

These categories share a common trait: a strong psychological characterisation. Indeed the most-followed accounts diverge from the tripartite regular distribution we had predefined to group different profiles: low scores (lower 25%), average scores (50%, median values), high scores (upper 25%). On the contrary, the most popular personalities on Twitter show values polarised toward the top or the bottom of each psychological trait.

Celebrities: atypical profiles

Results on emotional instability are particularly telling: low values of neuroticism are associated to all the above mentioned categories, apart from Twitter stars. Their generally straightforward and changing communication style might be paired wih a certain psychological frailty, or an inclination to experience negative emotions, as anger, anxiety or sense of guilt.

At the opposite, the Twitter accounts of companies, institutions and media oultets share minimal levels of emotional instability. This seems consistent with the public image they wish to promote and their adoption of sober and steady communication approaches.

Politicians show a hybrid profile, having affinities both with impersonal accounts and the showbusiness/communication macro-category. Openness to new ideas and propensity to plan but also a certain inclination to mood swings emerge from their tweets.

Finally, musicians and sportspeople show almost identical values of agreeableness and conscientiousness. This could be indicative of a very similar distribution of trust in other people and a results-oriented mindset in the two groups. A more detailed description of our work is available at