Tasks for Temporal Graph Visualisation

In [1], we describe the design and development of a task taxonomy for temporal graph visualisation. This paper details the full instantiation of that task taxonomy. Our task taxonomy is based on the Andrienko framework [2], which uses a systematic approach to develop a formal task framework for visual tasks specifically associated with Exploratory Data Analysis. The Andrienko framework is intended to be applicable to all types of data, however, it does not consider relational (graph) data. We therefore extended both their data model and task framework for temporal graph data, and instantiated the extended version to produce a comprehensive list of tasks of interest during exploratory analysis of temporal graph data. As expected, our instantiation of the framework resulted in a very large task list; with more than 144 variations of attribute based tasks alone, it is too large to fit in a standard journal paper, hence we provide the detailed listing in this document.

💡 Research Summary

The paper presents a comprehensive task taxonomy for the exploratory analysis of temporal graph data, extending the Andrienko framework—originally devised for time‑series and multivariate data—to accommodate relational structures. The authors begin by identifying a gap: while Andrienko’s systematic approach defines visual tasks such as lookup, comparison, pattern search, prediction, and summarisation, it does not address graphs where nodes and edges carry attributes that evolve over time. To fill this void, they redesign the underlying data model as a three‑dimensional construct: time × graph element (node or edge) × attribute, and they also introduce a “time‑structure” dimension for purely topological changes (e.g., subgraph appearance, path evolution, community dynamics).

Each dimension is mapped onto the basic Andrienko task types, yielding a matrix of possible queries. For attribute‑based tasks, the authors systematically vary three factors: (1) the nature of the attribute (quantitative vs. categorical), (2) the temporal scope (single instant, contiguous interval, or non‑contiguous set of instants), and (3) the comparison target (single node, a set of nodes, or the whole graph). By enumerating all permutations, they arrive at more than 144 concrete task specifications. Examples range from simple lookups—“What is the value of attribute X for node A at time t?”—to complex pattern searches—“Identify all nodes whose attribute X increases by more than 20 % between two user‑defined intervals and whose degree centrality simultaneously drops below a threshold.” This exhaustive list serves as a checklist for designers to ensure that a visualization system can support the full spectrum of analyst questions.

The taxonomy also covers structural tasks that focus on the evolution of the graph’s topology. These include subgraph matching across time, tracking the shortest path between two nodes as edges appear or disappear, detecting the birth and death of communities, and measuring changes in connectivity metrics (e.g., clustering coefficient, betweenness). Such tasks are formalised as “temporal‑structure” queries, for instance: “Does a previously existing clique persist across the interval