From Texts to Networks: Detecting and Managing the Impact of Methodological Choices for Extracting Network Data from Text Data
- Jana Diesner
- KI - Künstliche Intelligenz
Documentaries are meant to tell a story, i.e. to create memory, imagination and sharing (Rose, 2012). More specifically, the goal with social justice documentaries is to motivate change in people's knowledge and/ or behavior (Barrett & Leddy, 2008). How can we know if a production has achieved these goals? And how early in the life cycle of a film project can we answer this question? These research questions are of high practical relevance: Given the scarce resources for financing documentaries, major funders such as the Sundance Documentary Fund, the Ford Foundation and BritDoc have a strong need for reliable, comprehensive and efficient assessments of the return of their investment, where their target function with these investments is to cause change on a group, organizational or societal level (Clark & Abrash, 2011; KnightFoundation, 2011). Additionally, filmmakers and (impact) producers are interested in tracking the effectiveness of the outreach and campaign work around a production from the earliest stages of a film project on. They can leverage this information to strategically allocate resources and tapping into existing social capital and awareness for an issue. We report on our research on developing, applying and evaluating a theoretically-grounded, computational solution for the practical assessment of the impact of documentaries in a scalable, empirical and systematic fashion. We approach this problem from a network analytical perspective: based on the assumption that documentaries are produced, screened and watched as part of larger and continuously changing ecosystems that involve multiple stakeholders and the (potential) flow of information between them, we track, map, fuse and analyze social and semantic networks that represent this information; using data from media, social media and focus group interviews. More specifically, we leverage techniques from social network analysis and natural language processing to detect and assess the structure, functioning and dynamics of the web of social agents and information associated with (the main issue of) a movie.