{#- vim: set ft=htmldjango : -#}

The explanations provided in the present documentation as well as in the info texts in the visualization, use terminology and concepts from information visualization. We strove to make these texts as accessible as possible to a wider audience, but some basic understanding of the concepts used is still required. Here, we introduce the terms we used, and explain the visualization types.

Interactive Visualization

Card et al. formalized the so-called data visualization pipeline in 1998 (see here for reference). This pipeline specifies the steps that data goes through, from its source representation to the point where it is presented to the users using the visualization. The pipeline consists of four steps, in which certain operations are applied, transforming the data for the next step. After the last step, the data is present as pixels on a screen (or as ink on paper). The pipeline allows for viewers to interact with and influence the process during each of the four steps, changing the end result in different ways. In a first step, the data has been processed into a consistent, clean form, which is stored in the database. In general, user interaction only applies to the last three steps, and we will focus on these in what follows.

In a second step, through data transformations, the source data is transformed into data tables (i.e., suitably structured data). These transformations can include general data mapping, aggregation (such as counting or averaging), and filtering (e.g., include data after the year 1000 CE only). By applying filters in the visualization or changing, for example, the display mode in the settings (cf. {{ full_documentation_link('settings-display-mode', 'here') }}), viewers influence this second step of data transformation.

In the third step, called the visual mapping, the data tables are mapped to visual structures. For example, in a bar chart visualization, data items are mapped to rectangles, and the value is mapped to the height of the respective rectangle. In general, data attributes are mapped to visual variables. Visual variables include, but are not limited to: position, length (height, width, diameter), area, shape, color value, color hue, or texture. Examples for interaction with regard to the visual mappings are to switch between qualitative and quantitative mode for the timeline (cf. {{ full_documentation_link('settings-timeline-mode', 'here') }}). In this case, data is either mapped to rectangles of different colors, or to stacked area. Note: In this example, the data transformation step is also affected.

In the fourth and last step, the visual structures are then rendered to the views. In this step, the view transformations, the visual perspective on the data is also modified. This is done via geometric transformations: translation, scaling, and rotation (although the latter is used less frequently). Viewer interaction concerning this step includes, for example, zooming and panning (cf. {{ full_documentation_link('terminology-zooming-panning', 'here', true) }}).

Zooming and Panning

Zooming and panning are two types of interaction taking place in the image space of the visualization. However, they possibly interact with other parts of the information visualization pipeline, rather than just with the view transformation step.

Zooming

Zooming is the process of increasing or decreasing the scale of the visualized image. In a map, this means showing a smaller area of the map in more detail (zooming in), or showing a larger area in less detail (zooming out). In a timeline, this could mean showing a shorter time span in more detail, or a longer time span in less detail.

In most cases in information visualization, zooming is not merely a geometrical scaling operation (such as zooming in on a picture, simply enlarging the size in which pixels are shown). Rather, zooming in means that data can be displayed with more detail and less aggregated. Similarly, when zooming out, data needs to be aggregated more. Hence, zooming often involves not only the view transformation step, but also the data transformation and visual mapping steps as parts in the information visualization pipeline. This type of zooming is also called semantic zooming, as opposed to geometric zooming, which only affects the view transformation step.

Panning

Panning is the process of changing the geometrical translation of the visualized image. Panning does not affect visual mapping, but only view transformation. Examples for panning include:

Multiple Coordinated Views

{% if show_figures %}
Screenshot of the visualization, consisting of multiple coordinated views. Each view visualizes one aspect of the data, and interactions with one view are reflected in the others.
{% endif %}

A multiple coordinated views (MCV) visualization consists, as the name implies, of multiple views. These views are visually separated, either just by empty space between them or by borders. In Damast (see {{ full_documentation_link('fig-full-vis-screenshot', 'this screenshot', true) }}), each view is displayed in a separate user interface (UI) element. These UI elements are called panes and can be resized, rearranged, or maximized, similar to the way windows can be interacted with in an operating system such as Microsoft Windows.

In an MCV visualization, each view shows a different perspective on the same data; that is, the view visualizes a specific aspect of the data. In Damast, one view visualizes the temporal aspect of the data (the timeline), another view the geospatial aspect (the map), and so on. However, the same underlying data (pieces of evidence) is shown in each view. Further, the views are coordinated, meaning that interaction with one of the views is reflected by changes in other views. {{ full_documentation_link('fig-full-vis-screenshot', 'This screenshot', true) }} provides an example: Filtering by time range in the timeline view also affects the data shown in the map view. After filtering, only places with evidence from that time range are shown. For more details on the different interactions; see the sections on {{ full_documentation_link('terminology-filtering', 'filtering', true) }}, {{ full_documentation_link('terminology-selection', 'selection', true) }}, and {{ full_documentation_link('terminology-brushing-linking', 'brushing and linking', true) }}.

Filtering

Damast is a top-down visualization, meaning that initially, all data is shown, and users can then drill down into that data to find smaller, more specific subsets of the data. The drill-down is realized by applying filters to the data. A filter decides for each datum whether it matches specific criteria or not. Applying a filter to a dataset results in a subset of the dataset (not necessarily a proper subset in mathematical terms).

An example for a filter in Damast is to select a time span from the timeline. The filter then specifies the time span within which the evidence must lie to still be visualized. Another example is to specify one or more sources that the evidence must be attributed to; in this case, the data visualized stems from these sources only.

Our MCV visualization ({{full_documentation_link('terminology-mcv', 'see above', true) }}) implements multi-faceted filtering. That means that each view can have a separate filter active at the same time. Because the views show different aspects of the same data, the filters, too, apply to different aspects of the data: The timeline filter applies to the temporal aspect of the evidence, the map filter applies to the geospatial aspect of the evidence, and so on.

Generally, one should be aware of how these filters work between as opposed to within views. Between views, the filters are applied in conjuction; that is, a piece of evidence is shown only if it matches all filters. For example, if the map and timeline both have an active filter, evidence is only visualized if it is within the specified time span and within the specified geographical region.

Within views, the filters are applied in disjunction; that is, a piece of evidence matches the filter if it matches any of the criteria. For example, if a religion filter with two religions is active, it matches evidence that has either one or the other religion. This behavior is logical, in that there can be no evidence that has both religions at the same time, or is attributed to two places at once.

Selection

Selection is a user interaction with the data. In Damast, selection is done by clicking on some visual element with the computer mouse. For example, clicking a glyph (i.e., a symbol representing one or more places) in the map informs the visualization that the user selected that glyph. Note that what the visualization does in reaction to clicking is no longer part of the selection itself. For the main purpose of selection, see {{ full_documentation_link('terminology-brushing-linking', 'brushing and linking', true) }} below.

Brushing and Linking

{% if show_figures %}
Example for brushing and linking. Evidence of two religious groups is visualized (top). After selecting the Church of the East (COE) in the religion view, evidence of COE is brushed, and the respective visual representations of that data are linked in all views (bottom). In the religion view, all other religions' representations are desaturated and darkened. In the map, all map glyphs for places or clusters of places not containing COE evidence are darkened and desaturated.
{% endif %}

Brushing and linking is a process used in interactive MCV visualizations to help users understand the connection between different views, or rather, between different aspects of the visualized data. First, the user {{ full_documentation_link('terminology-selection', 'selects', true) }} some visual element in the visualization. The visualization interprets that selection, and performs brushing on the underlying data that is represented by that element. Next, the visualization links the brushed data in all views by applying specific highlighting to them.

In Damast, brushing always happens on a subset of the visualized evidence. The linking is then done in all views, including the one the selection and brushing originated from. In most views, we display linked elements by keeping their saturation, while all elements that are not linked are desaturated and darkened. A notable exception to this behavior is the timeline, where linking will show the temporal data from the brushed subset only, while all other data will be hidden.

Clicking on the same selected element again will revert the selection and also clear the brushing and linking. Selecting any other visual element will instead replace the selection and brushing and linking will apply to the new subset accordingly. In the map, the brushing and linking can also be cleared by clicking in an empty place.

A source of confusion with the term brushing can be that it is interpreted in the sense of paint brush; in that sense, the brushing itself would be a change of the visual representation of the elements (e.g., their being highlighted). However, this change of the visual representation is the linking part. Rather, brushing should be understood as in touching (or brushing) with one's fingers: Selecting a subset of data touches (brushes) them, and the subset of data is highlighted in all views, thereby visually linking the data to the selection and providing context.

{{ full_documentation_link('fig-brushing-linking', 'This screenshot', true) }} shows an example of brushing and linking in Damast: Evidence from two religions is visualized. Clicking on one of these religions in the religion view selects it and brushes the respective evidence. The visual representations of this subset of evidence is then linked in all views by desaturating the elements that are not part of the subset.

Visualization Types

A number of visualization types are used in Damast. The proper scientific terms are used in the description below, and are introduced here.

Bar Chart and Stacked Bar Chart

In a bar chart, categories are represented by rectangular bars, usually with a common baseline in one dimension, and a common width. The height of each bar encodes a value associated with the respective category. Bar charts can be horizontal as well, in which case the height of the bars is constant, and the width encodes value instead. In Damast, the bar charts used are horizontal, for example in the {{ full_documentation_link('source', 'source view', true) }}.

A special case of bar charts are stacked bar charts, in which each category, or bar, is further divided. Each segment of the bar encodes a sub-category's value. In Damast, stacked bar charts are used in the {{ full_documentation_link('untimed-data', 'untimed data view', true) }}, where each general religious affiliation is represented by a bar, and its specific religious groups by segments of that bar. In all data mode, all regular bar charts turn into stacked bar charts, with one segment for active data, and one for filtered-out data. Similarly, in confidence mode, a segment for each represented level of confidence is shown.

An even more special case is the normalized stacked bar chart, where the width (or height) of the bars is constant across bars as well. Hence, the width of individual segments encodes their relative proportion within the parent category. Normalized stacked bar charts appear in the {{ full_documentation_link('source', 'source view', true) }}, where they show the distribution of religious groups (or confidence levels) within each source.

Stacked Histogram

A histogram shows aggregated values of one value in dependence of another value, usually time, which is split up into bins. It is similar to a line chart, but the area between the line and the x-axis is filled in. A stacked histogram, similar to a stacked bar chart, is a chart where multiple such areas are stacked on top of each other for each bin. This representation makes it harder to read individual values, but provides a better understanding of the sum over all categories and general trends. In Damast, stacked histograms are used in the quantitative mode of the {{ full_documentation_link('timeline', 'timeline view', true) }}.

Indented Tree

An indented tree is a visualization for hierarchies. Nodes of the hierarchy are represented as elements placed in individual rows, or columns. To signify parent-child relationships, children are indented further than their parents. Indented trees are often used in file managers to show directory structures, or in mail programs to display e-mail threads. In Damast, an indented tree visualization is used to show the {{ full_documentation_link('religion-hierarchy', 'hierarchy of religious groups', true) }}.