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

4 Design

FlowSampler is a visual analytics system that comprises of four visualisations components each highlighting a different attribute in the data. We have chosen a visual analytics approach, as the task we address is exploratory and requires human judgement for pattern analysis and evaluation. By providing a visual interface, spatial planners can establish and fine tune their analytical procedure, identify uncertainty and biases in the data, and communicate their findings in an interactive visual environment. Our system should allow spatial planners to identify significant flow pathways that connect various locations in a given territory. Specifically, we are interested in

(1) flow patterns that exhibit characteristics of routine behaviour and (2) sequence of movements that can be used to characterise spontaneous, unexpected, behaviour. Our model considers a flow path to be routine if it contains a large number of trajectories that are made by few people. Conversely, a flow pathway is considered to exhibit spontaneous characteristics when it is infrequently traversed by a large number of people (fig. 1).

Fig. 1. Model to distinguish between routine and spontaneous characteristics in flow pathways

4.1 Data Transformation

To model urban flows (fig. 1), we propose a data transformation procedure for which we can transform spatial point type data into a graph structure suitable for expressing flow pathways [40]. The procedure consists of two steps. We begin by discretising the territory with an n2 grid where nn possible trajectories may occur. Next, we propagate

a directed graph G(V, E), where nodes vi ∈ V are cells in the grid generated in step 1. A directed edge E(i, j) represents a movement trajectory from node vi to node vj if a tweet has been made by a user in cell vi and cell vj in chronological order. We adopt two attributes as edge weights: The number of aggregated trajectories T(i,j) and the number of unique people P(i,j) that move between cell vi and cell vj. Looping or self directed edges, for instance E(i, i), are also accounted for in the same manner.

4.2 Interface Components

Our interface comprises of four components in a linked, integrated view (fig. 2). It contains a flow map, a time selection widget, a trajectory selection widget and a headcount selection weight. We implemented dynamic zooming in the flow map. Both weight and time selection widgets have dynamic filter ranges for the user to define filter boundaries. Inline with the visual information seeking mantra [41], all filter ranges are set to span the entire distribution while the map is set to the furthest zoom level on initialisation. This provides spatial planners with an overview of the data before further analytical task are carried out. The zoom level of the flow map can be modified with the mouse wheel and the filter range by manipulated by moving the interactive range sliders or selecting individual bars.

Fig. 2. FlowSampler interface components. (A) Flow map with flow pathways represented as arcs. Red arcs represent incoming flows while blue arcs represent outgoing flows. (B) Time selection widget. (C) Trajectory selection widget (D) Headcount selection widget.

Fig. 3. Comparison between two flow representations. (A) Depending on the physical geometry of the territory and flow patterns in the data, polyline representations may create distracting crossings that are confusing to interpret. (B) Arc representations address this problem by separating the short flow from the long flows.

Fig. 4. FlowSampler configured to show flow pathways at two levels of spatial-temporal resolution. (A) Coarse granularity reveals general trends in the data. (B) Fine granularity allows spatial planers to identify outliers and nuances such as movements between locations that only occur at specific hours of the day.

Flow Map. The flow map provides a spatial view of the data. The principle behind flow map is based on a node-link type representation where trajectories are plotted as lines that link origin to destination. While this approach produces maps that are familiar to many people, it does not scale well to large numbers of trajectories. To reduce visual clutter, flow maps merge trajectories that share similar origins and destinations. A line of varying thickness is then use to express the number of trajectories that have been aggregated. Similarly, ellipses of varying diameter are used to represent selfdirected flows. We adopt a node-link representation that can be super imposed onto a variety of base maps depending on various planning needs. To reduce ambiguity during interpretation, we represent flows with tapered polylines as recommended in literature [42]. While polylines are effective when flow distances are relatively short, they can become problematic when connections between two distant cells create long diagonal lines that may cause overlap or distracting crossings. To address the problem, we curve the polyline to form an arc. Arcs were chosen for the alternative representation, as they are a computationally cheap solution to avoid distracting crossings by separating the short flows from long flows. A disadvantage of the arc representation however is the added visual complexity it introduces to the display. There are two interactive features that support data exploration. Spatial planners may filter the flow to focus on a specific range of flow pathways in the data or select a cell to highlight the incoming and outgoing flows related to it. We tint selected flows with divergent colours to emphasize directionality.

Time Selection Widget. The time selection widget is an interactive split bar chart where each bar represents a time unit (i.e. week/day/hour) predefined by the spatial planer. Bars are arranged in a chronological order. The height of each bar in the upper half of the bar chart is used to encode the number of trips that occurred during that period while the lower half of the bar chart is used to indicate the total number of flows related to a selected cell. The lower half of the bar chart will be empty if no cells are selected.

Trajectory and Headcount Selection Widget. The trajectory selection widget and the headcount selection widget are histograms that visualise the distribution of the aggregated trajectories and unique people who travelled along a certain flow pathway respectively. The intervals of the histogram are determined by a linear interpolation by default but spatial planners may dynamically switch the display to a logarithmic interpolation in the event of a highly skewed distribution.

 
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
 
Subjects
Accounting
Business & Finance
Communication
Computer Science
Economics
Education
Engineering
Environment
Geography
Health
History
Language & Literature
Law
Management
Marketing
Philosophy
Political science
Psychology
Religion
Sociology
Travel