Characterizing place: an empirical comparison between user-generated content and freelisting data

Abstract

GIScience 2016 Short Paper Proceedings Characterizing place: an empirical comparison between user-generated content and free-listing data F. M. Wartmann 1 , C. Derungs 1,2 , R.S. Purves 1,2 Department of Geography, University of Zurich, Winterthurerstrasse 190, CH-8057 Zurich, Switzerland Email: {flurina.wartmann; curdin.derungs; ross.purves}@geo.uzh.ch URPP Language and Space, University of Zurich, Freiestrasse 6, CH-8032 Zurich Abstract Methods to gather information from the public about place range from ethnographic approaches such as free listing to automatic extraction from user-generated content. We compared aspects of place (location, locale and sense of place) contained in free lists elicited from participants recruited on site with tags from georeferenced Flickr images. Using manual annotation we assigned content as toponyms (location), landscape elements (locale) and feelings (sense of place). Flickr tags contained more toponyms than free-listing data, but almost no information relating to feelings. Landscape elements were prominent in both data sets, with those captured by free lists and Flickr being cognitively more salient than those only captured by free listing, suggesting they represent basic levels. In Flickr, landscape elements consisted of basic levels in different languages (e.g. mountain, Berg), while free lists contained landscape elements both at the basic and subordinate level (e.g. Arvenwalder, arolla pine forests). We conclude that both methods yielded information about locale, with Flickr contributing basic levels and free lists also more detailed information, but that Flickr provided little information about sense of place compared to in situ free-listing elicitation with participants. 1. Introduction The importance of taking into account meanings people assign to places in management and planning is increasingly being recognized (Jones 2007; Prieur et al. 2006). However, collecting such information from the public is non-trivial, and different approaches are used, ranging from ethnographic work to extraction from user-generated content. Given the diversity of approaches, it is crucial to compare different approaches and understand what types of information are captured by different methods such that research on place can effectively capture relevant information. In this paper, we compare free-listing elicitation with automated extraction of place descriptions from user-generated content. In free-listing experiments, participants are asked to freely name examples of, for instance, categories they associate with the landscape they currently find themselves in (Bieling et al. 2014; Wartmann et al. 2015). Another method increasingly used is to automatically extract information from user-generated content, often in the form of image tags (e.g. Jenkins et al. 2016; Purves et al. 2011; Rattenbury and Naaman 2009). We compared these two methods based on the information about place they contained. 2. Methods We chose two sites in the Swiss Alps (Figure 1) where outdoor free listings on landscape categories had been conducted with 59 participants recruited on site (Wartmann et al. 2015). Terms were elicited with a Swiss German question that literally translates to: ‘what is there for you in a