Investigating Linguistic Indicators of Generative Content in Enterprise Social Media

Teamwork is at the heart of most organizations today. Given increased pressures for organizations to be flexible, and adaptable, teams are organizing in novel ways, using novel technologies to be increasingly agile. One of these technologies that are increasingly used by distributed teams is Enterprise Social Media (ESM): web-based applications utilized by organizations for enabling communication and collaboration between distributed employees. ESM feature unique affordances that facilitate collaboration, including interactions that are generative: group conversations that entail the creation of innovative concepts and resolutions. These types of interactions are an important attraction for companies deciding to implement ESM. There is a unique opportunity offered for researchers in the field of HCI to study such generative interactions, as all contributions to an ESM platform are made visible, and therefore are available for analysis. Our goal in this preliminary study is to understand the nature of group generative interactions through their linguistic indicators. In this study, we utilize data from an ESM platform used by a multinational organization. Using a 1% sub sample of all logged group interactions, we apply machine-learning to classify text as generative or non-generative and extract the linguistic antecedents for the classified generative content. Our results show a promising method for investigating the linguistic indicators of generative content and provide a proof of concept for investigating group interactions in unobtrusive ways. Additionally, our results would also be able to provide an analytics tool for managers to measure the extent to which text-based tools, such as ESM, effectively nudge employees towards generative behaviors.