#WhoAmI in 160 Characters? Classifying Social Identities Based on Twitter Profile Descriptions

Abstract

We combine social theory and NLP methods to classify English-speaking Twitter users' online social identity in profile descriptions. We conduct two text classification experiments. In Experiment 1 we use a 5-category online social identity classification based on identity and self-categorization theories. While we are able to automatically classify two identity categories (Relational and Occupational), automatic classification of the other three identities (Political, Ethnic/religious and Stigmatized) is challenging. In Experiment 2 we test a merger of such identities based on theoretical arguments. We find that by combining these identities we can improve the predictive performance of the classifiers in the experiment. Our study shows how social theory can be used to guide NLP methods, and how such methods provide input to revisit traditional social theory that is strongly consolidated in offline settings.

Stepping Out of the Ivory Tower for Ocean Literacy

“…The accuracy of our keyword-based user classification scheme was determined by manually checking the profiles of a random subsample (∼10%) of users, from which the assigned classification was either verified or not. In our subsample, the classification scheme returned a precision rate of 85% -a success rate in line or better than similar studies that conducted user classification analysis on Twitter (Wagner et al, 2012;Barthel et al, 2015;Priante et al, 2016;Côté and Darling, 2018b;Haustein, 2018). The simple dichotomous categorization scheme greatly reduces the risk of misclassification, however, the algorithm is not without limitations.…”

“…Biographical user-related data provides one of the most accurate depictions of a user's true identity (Wagner et al, 2012), and was therefore the primary basis for our classification scheme. Rooted in previous methodologies (Barthel et al, 2015;Priante et al, 2016;Côté and Darling, 2018a,b), we built a pre-defined keyword and expression string list using words that related to our INREACH classification group, and were therefore likely to be mentioned in a user's biographical data ( Table 1). The "stringr" package (Wickham, 2017) in R was used to filter user profiles that contained relevant keywords and expressions.…”

Towards Real-Time, Country-Level Location Classification of Worldwide Tweets

“…Re-1. http://www.bbc.co.uk/programmes/b04p59vr searchers have attempted to infer attributes of Twitter users such as age [23], [46], gender [6], [31], [35], [46], political orientation [11], [12], [40], [41] or a range of social identities [44]. Digging more deeply into the demographics of Twitter users, other researchers have attempted to infer socioeconomic demographics such as occupational class [42], income [43] and socioeconomic status [28].…”

Bots and online hate during the COVID-19 pandemic: case studies in the United States and the Philippines

“…Each identity term in the lexicon was further subdivided into four classes: political identities (e.g., senator, president), gendered identities (e.g., women, transgender), racial/ nationality identities (e.g., Black, Filipino), and religious identities (e.g., priest, imam). Subclasses were based on related work on hate speech and social media identity prediction more generally [46,65]. Acknowledging that identities are intersectional [28], this subcategorization scheme was set up such that each identity term could be assigned any combination of the four identities mentioned above.…”