Log in / Register
Home arrow Computer Science arrow Social Informatics
< Prev   CONTENTS   Next >

6 Limitations, Uncertainty and Bias

We have described FlowSampler, a visual analytics system that supports the extraction and exploration of urban flows in geolocated social media data. A key advantage of this system is that it enables planners to interactively reconfigure the interface to explore and detect patterns in the data at various spatial and temporal granularities. Furthermore, our system allows spatial planners to include external geographic information in form of base maps to evaluate the significance of patterns that they have identified. To show the functionality and scalability of our system, we presented two use cases that investigate urban flows at different spatial and temporal scales. We identified pathways of routine movement that occur within Flanders, a region in Belgium, over the duration of a year, and traced exceptional transit activity converging on and subsequently occur within Cilento, a national park in Italy over the touristic season spanning three months. While initial deployment of FlowSampler with the spatial planners in our department has resulted in positive feedback, several discussion topics have been raised.

Skewed Demographic. As existing studies indicate that majority of the online social media users are young adults [43], there is concern that the flow patterns we detect only represent a partial slice of the actual population on the ground. While we acknowledge this limitation, we would like to point out that our approach provides equally valuable and alternative insights that are complimentary to the results derived from other urban flow analysis techniques.

Sporadic Activity. We observe a non-linear distribution of tweeting activity in the form of a long tail where a handful of highly active users are trailed by a substantially larger number of people who tweet sporadically. Because highly active users have trajectories that comprise of many more trips than sporadic user, the uneven distribution implies that certain movement pathways will be over emphasized thus skewing the overall representation. We address this challenge by allowing spatial planners to interactively modify the flow map to determine which attribute edge thickness encodes (i.e. the number of trips or the number of people). This facilitates visual comparison between both attributes in order to identify bias in the representation. Another feasible solution is to pre-filter overly active users to remove the bias entirely from the analysis however this narrows the slice of the population being studied.

Privacy. Our system is designed to present information about aggregated movement behavior yet we provide functionalities for the information to be disaggregated. While we acknowledge that it maybe difficult to prevent the recovery of personal information under such circumstances, imposing control measures to displace or distort the data maybe counter productive for the spatial planners.

Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
Business & Finance
Computer Science
Language & Literature
Political science