NECSUS (Jan 2023)

Screening the financial crisis: A case study for ontology-based film analytical video annotations

  • Jan-Hendrik Bakels,
  • Matthias Grotkopp,
  • Thomas J. J. Scherer,
  • Jasper Stratil

DOI
https://doi.org/10.25969/mediarep/21732
Journal volume & issue
Vol. 12, no. 2
pp. 386 – 407

Abstract

Read online

This paper presents a dataset of fine-grained film analytical annotations (Melgar and Estrada and Koolen 2018) for a corpus study of feature films, documentaries, and television news on the Global Financial Crisis (2007-), generated by the research group Affektrhetoriken des Audiovisuellen (Freie Universität Berlin and Hasso-Plattner-Institute Potsdam, 2016-2021, see Bakels et. al. 2020a). The semantic video annotations are based on the AdA Filmontology (v1.8), which consists of eight annotation levels, 78 annotation types, and 501 annotation values (Bakels et. al. 2020b). Each level, type, and value has a unique resource identifier (URI) as well as an English and German name and description. In our paper, we reflect on the specific challenges of capturing film-analytical claims of embodied viewing experiences in an ontology-based taxonomy. We further critically discuss aspects such as intercoder-reliability, consistency, as well as the requirements of training and synchronising expert annotators. The dataset contains more than 92,000 manual and semi-automatic annotations authored in the open-source-software Advene (Aubert/Prié 2005) by expert annotators, as well as more than 400,000 automatically-generated annotations for wider corpus exploration. The annotations are published as Linked Open Data under the CC BY-SA 3.0 licence and available as rdf triples in ttl files and in Advene’s non-proprietary azp-file format, which allows instant access through the graphical interface of the software. Via a web application all annotations can be downloaded, queried, and visualised in conjunction with password-protected access to the source video files (Agt-Rickauer 2022). This dataset is of interest for research on the financial crisis discourse or the specific films and broadcasts; also, the dataset serves as a proof of concept for ontology-based video annotation and as a provider of training data on film analytical concepts such as shot length, camera movements, or affective tonalities.

Keywords