Twitting agencies: reputational signals of regulatory agencies in social media

Public agencies’ Twitter is becoming a mainstay in organizational communication and strategic communication strategies, being the most used social media application in strategic communication campaigns, official public relations, advertising, and marketing campaigns. Government agencies are increasingly using Twitter to serve a myriad of organizational communication needs (Waters and Williams 2011). More than 1000 agency, department, initiative, or team Twitter accounts exist within the US federal government alone (Zavattaro and Sementelli 2014). However, social media is not as widely adopted as other forms of information and communication technologies but are used as additional and oftentimes parallel communication channels to existing channels (such as websites) (Mergel 2012). In addition, Twitter is a proper channel for communication strategies that agencies can tailor to their reputational needs. Indeed, Twitter offers companies an influential environment in which to enhance their reputation and build relations with existing and potential clients (Page 2014). Previous research has already revealed that organizations are unlikely to become interactive on social media platforms such as Twitter unless they are called upon to protect their organizational reputations in the wake of an institutional crisis (Coombs 2007). Otherwise, organizations prefer to engage in one‐way communication strategies and minimally respond to publicly posted messages on the web (Cooper and Owen 2007; Glenny 2008). Assuming that Twitter is a proper channel for communication strategies that public agencies can tailor to their reputational needs, in this study we compare the reputation signals three regulatory agencies of communication (United Kingdom`s Ofcom; Portugal`s Anacom and Brazil`s Anatel) adopt. We collected tweets from agencies` Twitter profile through Twitter API. We collected the maximum volume allowed by the standard version of the API [], so the spans for each agency depend on their use of TWitter. In order to perform supervised learning over each agency` Twitter dataset with respect to reputation signals, we randomly selected a sample corresponding to 9.96% of the whole dataset. Then, the selected sample was presented to a coder; the coder indicated whether a tweet contained content related to Carpenter`s reputations signals (moral, procedural, technical or performative). From the coded tweets, we started to build a classifier. Each tweet is a document. After removing stop words, very frequent terms (present in more than 90% documents) and very infrequent terms (present in less than 5% of the documents), we built a matrix of TF-IDF features, and conducted experiments over twelve different classifiers. For each classifier, a 10-fold cross validation was performed, allowing to choose the best classifier. After choosing the best classifier for each agency, we run the model training on 66% of the dataset and testing on 33%. Finally, we label all agency tweets according to reputation signals using the built supervised model. Our results allow to analyze agencies` tweets according to their engagement, reputation signals and evolution over time. Our preliminary analysis reveals striking differences among regulatory agencies in their use of Twitter, indicating theoretically similar regulatory agencies adopt very different communication strategies in social media.