Anika Christen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Anika Christen has a total of 3 co-authors. Here are the top co-authors by number of shared publications:
- Anastasia Fiolka
Julius-Maximilians-Universität Würzburg, Germany
1 shared publication - Birgit Lugrin
Socially Interactive Agents, University of Würzburg, Würzburg, Germany
1 shared publication - Elisabeth Ganal
University of Würzburg
1 shared publication
Latest Publications
- Open Access
- 16 September 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-9
The Applicability of Using Virtual Social Contexts as Stimuli and Training Material for Social Context Research
Enabling artificial agents to be truly socially interactive and unobtrusive requires them to understand the social context of a situation. As a variety of different social context cues exist, both user studies and methods for context classification systems of agents are required to reliably categorize and recognise social contexts and respond appropriately and autonomously towards users. Compiling datasets from the real world can often be expensive, time-consuming and lack scalability. Thus, virtually simulated social contexts can be used to generate training data for context detection systems, but also stimuli material for user studies in a fast and cost-effective way. As a proof of concept, we introduce virtual social contexts and test their applicability for social context research with objective and subjective measurements. Therefore, we first implement virtual social contexts to compile our own dataset1. With subjective participant ratings, we evaluate the applicability of virtual social contexts as stimuli material for user studies. For this, we replicate a previous study with recordings of real social contexts using recordings of virtual social contexts instead. For our objective measurements, we use the computer vision models of YOLOv8 on object detection and pose estimation to train a classification model on our virtual image dataset and evaluate it on an image dataset of real social contexts. The results of the subjective and objective analyses indicate, that virtually generated social contexts can be used both for training classification systems and as stimuli material in user studies.
- 16 September 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-9
Open AccessThe Applicability of Using Virtual Social Contexts as Stimuli and Training Material for Social Context Research
Enabling artificial agents to be truly socially interactive and unobtrusive requires them to understand the social context of a situation. As a variety of different social context cues exist, both user studies and methods for context classification systems of agents are required to reliably categorize and recognise social contexts and respond appropriately and autonomously towards users. Compiling datasets from the real world can often be expensive, time-consuming and lack scalability. Thus, virtually simulated social contexts can be used to generate training data for context detection systems, but also stimuli material for user studies in a fast and cost-effective way. As a proof of concept, we introduce virtual social contexts and test their applicability for social context research with objective and subjective measurements. Therefore, we first implement virtual social contexts to compile our own dataset1. With subjective participant ratings, we evaluate the applicability of virtual social contexts as stimuli material for user studies. For this, we replicate a previous study with recordings of real social contexts using recordings of virtual social contexts instead. For our objective measurements, we use the computer vision models of YOLOv8 on object detection and pose estimation to train a classification model on our virtual image dataset and evaluate it on an image dataset of real social contexts. The results of the subjective and objective analyses indicate, that virtually generated social contexts can be used both for training classification systems and as stimuli material in user studies.