From the perspective of our research on geographic recommender systems, the goal of designing \systems and services that can understand `place' in a way we humans do" seems little ambitious and potentially misleading. We argue that we should aim at designing services which are, in a speci c sense, better than humans at understanding place. Every individual belongs to one or more social groups and, in general, the ability of humans to understand place conceptualizations from other social groups is rather limited. Geographic recommender systems should outperform humans in the handling of multiple group-speci c place models (Matyas & Schlieder, 2009; Schlieder & Kremer, 2012, accepted). One important consequence of looking at di erences in conceptualizations has been pointed out by Schlieder and Henrich (2011). The classical membership problem of place research { does the point X belong to place P { transforms into more complex problems: does user U believe X belongs to P? Do users U and V share similar beliefs about X belonging to P? While this statement seems trivial according to our everyday life experience, its implications on geographic recommending are widely overlooked (Winter et al., 2009). Our talk illustrates the need for more complex place models by analyzing data about touristic conceptualizations of urban spaces derived from a recent GPS tracking study of touristic exploration behavior. The data set is based on behavioral data from 17 rst time visitors to the town of Bamberg, Germany, who came for a one-day-trip and volunteered to participate in the study. They were handed a camera equipped with GPS and a magnetic compass, as well as a second GPS receiver with better positional accuracy for recording the track data. The participants were told to explore the town in whatever way they liked and for how long as they pleased. A rst analysis of the data is based on two dependent variables, the number of photographs taken in di erent regions of interest and the time the visitor spent there. We describe two marginal return models for place popularity, one based on photograph frequency, the other based on visit time. Individual di erences in spatial choices are compared using these models. A rst group of ndings relates to individual di erences in the popularity of places while a second group of ndings concerns the exploration strategies employed by the visitors. Within the former, we found that time-based and photograph-based measurements of interest in a place may signi cantly di er. We hold that this re ects a place's a ordance for tourist activities. The popularity of some places is linked to their visual attractiveness while other places lend themselves for dwelling. In the analysis of the data, we use extended periods of zero motion speed as an indicator for dwelling behavior and the photographing
activity as an indicator for touristic attention. We found that tourists which move along nearly the same track, are likely to photograph di erent sights. Conversely, people that happen to catch the same glance will add semantics to it only according to their speci c background, interests or even other places visited before. Based on our ndings, we suggest the following (partial) answer to the workshop challenge of automatically detecting the location of things based on behavioral data: { At least for most touristic places, there is no universally accepted location of the place that could be determined by machines (or by humans). Di erent place conceptualizations tend to coexist. The town of Bamberg, for instance, can be conceptualized as the beer capital of Bavaria or as the Unesco world heritage site of Baroque architectur or in many other ways. { However, it is possible to analyze the spatial { and the thematic di erences { of place conceptualizations by looking at data from close monitoring studies involving GPS tracks and photographs taken by visitors. { We argue that the task of automatically detecting the location of a touristic place should not just map a place name onto a single geographic footprint, but rather on a set of footprints, di erent for di erent communities and di erent for di erent activities.