April 22, 2021 View Manipulation ▪ Why to manipulate and change the view? ▪ datasets might be too large to show everything at once - reduce complexity of single view ▪ single static view might lead to visual clutter ▪ How to manipulate/change a view over time? ▪ select specific elements (items or attributes) ▪ reordering (sorting) of items - find patterns by ordering based on different attributes ▪ change parameters of a particular idiom - e.g. range of possible mark sizes ▪ semantic zooming ▪ switch between idioms ▪ …
April 22, 2021 Animated Transitions Example ▪ Maintain a sense of context between two states Animated Transitions What(Data) Compound network. How (Manipulate) Change with animated transition. Navigation between aggregation levels.
April 22, 2021 Element Selection ▪ Different design choices for element selection ▪ which elements can be selection targets? - data items, links, data attributes, levels within a data attribute, … ▪ one kind of selection vs. multiple kinds of selection (e.g.via hover) - multiple mouse buttons or combination with key presses for more advanced types of selections ▪ selection of single elements vs. selection of many elements ▪ selection of primary and secondary target - e.g. for path traversal from source to target in a directed graph ▪ Selection often defines the target of a next action
April 22, 2021 Selection Highlighting ▪ Provide immediate visual feedback to users about element selection ▪ different possibilities for highlighting of data items - changing colour (hue, luminance or saturation) for visual popout - add or change existing outline - change the size of a data item - motion coding (e.g. slightly moving items of moving pattern) ▪ different possibilities for highlighting link marks - changing colour - changing linewidth, shape (e.g. dashed) - … ▪ multiple highlighting design choices can be combined ▪ selected items might be connected via explicit visual links (connection marks)
April 22, 2021 Context-preserving Visual Links Context-preserving Visual Links What(Data) Any data. How(Encode) Any encoding. Highlight with link marks connecting items across views. How (Manipulate) Select any element. How (Coordinate) Juxtaposed multiple views.
April 22, 2021 Navigate: Changing Viewpoint ▪ Navigation can help to see a large and complex dataset from different points of view ▪ changing viewpoint of virtual camera changes the set of items visible in the camera frame ▪ often leads to a combination of filtering and aggregation ▪ Three main aspects of navigation ▪ zooming - moves camera closer (less items but with more details) or further away (more items but less details) from the image plane - geometric zooming vs. semantic zooming ▪ panning (translating) - moves camera parallel to the image plane (up and down or from side to side) ▪ rotating - spins camera around its axis (rarely used in 2D navigation)
April 22, 2021 Semantic Zooming ▪ In contrast to geometric zooming, the fundamental appearance of objects is no longer fixed ▪ object visualisation changes based on number of available pixels ▪ details added or removed based on the semantic zoom level ▪ different idioms might be used at different semantic zooms levels ▪ Constrained navigation limits the possible motion of the virtual camera ▪ avoids that user get lost by for example pointing the camera to an empty space or zooming out too much ▪ systems might also automatically compute the best viewpoint to view a selected item - smooth animated transition to the new viewpoint - powerful when combined with linked navigation between multiple views
April 22, 2021 Navigate: Reduce Attributes ▪ Number of attributes can be reduced in three different ways ▪ slice - single attribute value defines which items should be extracted - e.g. intuitive metaphor when reducing spatial data from 3D to 2D - possible to have higher dimensional slicing planes (hyperplanes) ▪ cut - plane dividing the viewing volume and everything on the side of the plane closer to camera viewpoint is not shown ▪ project - all items are shown but without the information for specific attributes - projections often used via multiple views • e.g. 2D views of a 3D XYZ scene (XY floor plan, YZ side view and XZ front view) • e.g. Mercator map projections from the surface of the earth to 2D maps
April 22, 2021 Reducing Items and Attributes … ▪ Reduction is one of the strategies for dealing with complexity in visualisations ▪ filtering eliminates elements - challenge: people might forget about the filtered elements ("out of sight, out of mind") ▪ aggregation combines many elements together - challenge: how and what to summarise (aggregate) in order to support a task (and match well with the dataset) ▪ filtering and aggregation can be applied to items or attributes ▪ Bidirectional operation ▪ reduce or increase the number of visible elements
April 22, 2021 Filtering ▪ Filtering often accomplished through dynamic queries ▪ tightly coupled loop between visual encoding and interaction ▪ e.g. user can interactively chose a range for the values of an attribute via graphical UI widgets ▪ Item filtering ▪ reduce number of items based on their values for specific attributes ▪ Attribute filtering ▪ keep number of items but reduce the number of shown attributes ▪ often used with attributes that can be ordered to filter out the low or high scoring ones ▪ Item filtering and attribute filtering can be combined
April 22, 2021 DOSFA Example ▪ Dimensional Ordering, Spacing and Filtering Approach (DOSFA) ▪ 215 attributes (representing word counts) and 298 points representing documents in the example Full dataset After filtering
April 22, 2021 Aggregation ▪ Group of elements represented by a derived element (aggregation) ▪ elements are merged rather than eliminated as with filtering ▪ challenge: aggregation (summary) might eliminate interesting signal in the dataset - e.g. see Anscombe's Quartet example presented earlier ▪ Item aggregation ▪ interactive aggregation and deaggregation of item sets ▪ Attribute aggregation ▪ group attributes by similarity measure and synthesize a new attribute based on average across the set ▪ dimensionality reduction (DR)
April 22, 2021 Histogram Example Histograms What(Data) Table: one quantitative value attribute. What (Derived) Derived table: one derived ordered key attribute (bin), one derived quantitative value attribute (item count per bin). How (Encode) Rectilinear Layout. Line mark with aligned position to express derived value attribute. Position: derived key attribute
April 22, 2021 Boxplot Charts Boxplot Charts What(Data) Table: many quantitative value attributes. What (Derived) Five quantitative attributes for each original attribute, representing its distribution. Why (Tasks) Characterise distribution; find outliers, extremes, averages; identify skew. How (Encode) One glyph per original attribute expressing derived attribute values using vertical spatial position, with 1D list alignment of glyphs into horizontally separated regions. How (Reduce) Item aggregation. Scale Items: unlimited. Attributes: dozens.
April 22, 2021 Spatial Aggregation ▪ Challenge in spatial aggregation is to take the spatial nature of aggregation into account when aggregating it ▪ changing the boundaries can lead to very different results → modifiable areal unit problem (MAUP) Central region with high density Central region with medium density Central region with low density
April 22, 2021 Dimensionality Reduction (DR) ▪ Preserve the meaningful structure of a dataset while using fewer attributes to represent the items ▪ assumes that there is hidden structure and redundancy in the original dataset ▪ multidimensional scaling (MDS) for more complex forms (not just a straightforward combination) of dimensionality reduction ▪ Dimensionally reduced data can be visualised as scatterplot (two attributes) or as scatterplot matrix (more than two attributes) ▪ only large clusters should be considered relevant ▪ fine-grained structure should not be considered reliable
April 22, 2021 DR for Document Collections Dimensionality Reduction for Document Collections What(Data) Text document collection. What (Derived) Table with 10'000 attributes. What (Derived) Table with two attributes. How (Encode) Scatterplot, coloured by conjectured clustering. How (Reduce) Attribute aggregation (dimensionality reduction) with multidimensional scaling (MDS) Scale Original attributes: 10'000. Derived attributes: two. Items: 100'000