The first step of this stage was to analyse the literature to identify the interactions in interactive visualisations for time-oriented data. This included the references surveyed as part of the view component and other references that were considered unsuitable for the view component, such as interactive three-dimensional visualisations. This step was used to filter out entries that do not contain any description about interactions, such as new visualisation techniques that were not tested in an interactive environment. For each of the initial 38 entries where interactions were described, the taxonomy of interaction techniques by Yi et al. (2007) guided the extraction of information. In addition to the seven basic categories that the authors describe –
select, explore, reconfigure, encode, abstract/elaborate, filter, connect – two additional
categories were used: visualisation configuration and design transformations. The first is discussed by Yi et al. as a category of interactions that are not exclusive to interactive visualisation, and thus their taxonomy does not include it. In the context of this thesis, however, it is important to consider aspects of visualisations that are part of the various stages of Card-Norman’s model, such as the parameters that guide the generation of visualisations. Visualisation configuration includes temporal transformations that are triggered by interface settings, such as resizing the visualisation window. Design transformations include temporal transformations that are applied in the generation of the visualisation, rather than being only applied by user input during runtime; this includes aggregations that are used to decrease the number of items on screen, for example. Each entry was analysed and textual descriptions were added to each category where appropriate, as seen in appendix D.
The second step was to filter the interactions that are relevant for the framework, which are those that belong to the following categories: reconfigure, abstract/elaborate, filter, visualisation configuration and design transformations. This excludes the select, explore and connect categories, which contain interactions that, for time visualisation, often involve purely visual transformations, such as highlighting elements, or transformations that involve other types of data, such as querying quantitative information related to certain times. These interactions are out of the scope of the framework (see chapter 9
for more discussion about the limitations of the framework). This step reduced the number of entries to 30.
The third step was to relate the textual descriptions to the properties of time that are modified or transformed by the interactions. Doing so required determining, from existing theories and models of time, which of these properties are important to support visual exploration. For this, the notion of a discrete bounded time domain and the concept of granularities defined by Bettini et al. (1998), described in chapter 2, was used. A time domain contains the temporal references that are part of any dataset being visualised; these are usually extracted from the data items being visualised and isolated from other non-temporal data. The discrete aspects refers to the fact that the temporal references are indivisible; in such a model, dividing 1 day requires changing that granularity that used for the time domain. In time visualisation, as the pixel is also an indivisible unit, generally speaking, it is sensible to reflect this characteristic in the abstraction of time as well. The bounded characteristic refers to the lower and
upper bounds, or the lower and upper time points, that define the extent of the time
domain. Although infinite time is conceptually possible, the fact that the visible visual
space is also limited makes it sensible to consider the limits of time where data items
cease to exist.
This representation of the time domain includes three other properties of granularity mappings: the unique granularity mappings (i.e. the mathematical properties of granularities), the types of labels and the variation of ordering of granularity labels. The types of labels are related to the different scales – changing between absolute time and relative time by re-labelling time points. The order of granularity labels is primarily related to cyclic time units, such as days of the week, in which the first day of the week can be changed; this is related to the variation of calendars and use of time in different application contexts.
Lastly, many visualisations subdivide the time domain into segments and display it in a non-contiguous manner. In order for the framework to support this, the concept of
segments is used, representing contiguous subdivisions of the time domain that can be
rearranged with varying length. Each reference was categorised by analysing these six
modifiable properties in relation to the textual description of interactions; table 6.1
displays the results of this.
The final step in designing the component was identifying categories of transformations based on these properties. This involved identifying groups of conceptually similar
Reference TE B G GL GO NS
André et al. (2007) ✓
Andrienko et al. (2011) ✓
Beard et al. (2008) ✓
Carlis and Konstan (1998) ✓ ✓
Cho et al. (2014) ✓
van der Corput and van Wijk (2017) ✓
Gad et al. (2015) ✓
Gschwandtner et al. (2011) ✓ Guerra-Gomez et al. (2013) ✓
Javed and Elmqvist (2013) ✓ ✓
Keim et al. (2004) ✓
Kothur et al. (2013) ✓ ✓
Krstajic et al. (2011) ✓
Lammarsch et al. (2009) ✓ McLachlan et al. (2008) ✓ Shen and Kwan-Liu (2008) ✓
Sips et al. (2012) ✓
Tominski and Schumann (2008) ✓ ✓
Tominski et al. (2012) ✓ ✓
Wang et al. (2009) ✓ ✓
Van Wijk and Van Selow (1999) ✓
Wongsuphasawat and Shneiderman (2009) ✓ ✓
Tableau1 ✓ ✓ ✓ ✓
Spotfire2 ✓ ✓
Table 6.1 Survey of the properties of time modified by interactions in the literature.
TE = Temporal Extent, B = Bounds, G = Granularity, GL = Granularity Label, GO = Granularity Order, NS = Number of Segments.
transformations and the various parameters that can be used with them. The properties of time were used to define three categories:
• Segmentation: operators in this category allow the number of segments to be modified by dividing the time domain, re-arranging segments or reversing the transformation;
• Granularity: operators in this category change the properties related to time granularities, by changing the level of aggregation, exploring a granularity struc- ture or modifying inner properties of granularities such as the order or labels of time points;
• Extent and bounds: operators in this category change either the temporal extent or the upper or lower bounds of the time domain.
The next section describes the operators within each category, including details of their functionality and the parameters that can be used.