Initial visitor models in a personalized museum guide
PaperCharles Callaway, Italy
When dynamically creating personalized sets of multimedia presentations in cultural heritage contexts such as museums, it is important to know as much as possible about new visitors while being minimally intrusive. Visitors are usually not interested in answering personal questions or frequently interrupting their visit to give feedback, while approaches to creating models of visitors such as stereotyping produce little initial variation and thus homogenize rather than personalize their experience.
More museums are including instrumented environments where physical sensors can continuously observe visitors, even before they begin using a system such as a museum guide. In these cases, it may be possible to create a user model by observing those users, and then using that data to predict either future behavior or a psychological profile. In the case of cultural heritage sites such as museums it is often useful to monitor the behavior of groups. Because group members arriving at museums already interact with each other before they even begin using a multimedia presentation system, if they have agreed to use a sensor-based museum guide it should be possible to create a more substantial user model based only on their initial group behavior before arriving at their first exhibit.
Ethnographic studies have shown that the length and quality of conversation within small groups of visitors in a museum can be seen as a fundamental indicator of successful engagement. If the structure or other characteristics of initial group conversation is predictive of the effect of a system’s personalization of multimedia presentations, we can better choose which adaptations will produce more conversation and interaction, and will thus lead to more engagement with the presentations, the museum’s contents, and the museum as a whole.
We illustrate this with a experiment based on the DramaTric system (Drama Tension Release by Inducing Conversation), an implemented mobile multimedia drama presentation system for groups. DramaTric is based on adaptive narration, and one of its uses of personalization is to select a variation in subsequent dramatic scenes depending on how much the group talked at the prior scene. DramaTric allows a group of visitors to move freely around a museum, while it produces a drama by coherently piecing together smaller drama segments. It consists of sensors both embedded in the museum and worn by the visitors, presentation devices, and software to implement dramatic presentations determined by sensor data.
A recent experiment showed that the adaptations in DramaTric had a significant effect on the amount of group conversation at the conclusion of its drama presentations. While that experiment analyzed changes in the amount of conversation over time, we were then interested if the same data could show whether the amount of initial conversation was correlated with later conversation. In particular, we look at three potential claims: 1) does the amount of conversation at the very beginning of the visit predict later conversation; 2) does the amount of conversation at a given presentation predict conversation at the next presentation; and 3) does conversation during a presentation predict conversation immediately after a presentation.