Conclusion - Data Warehouse Systems: Design and Implementation - page 576

them, which has a collaboration frequency attribute x . For every conference
in every year, we may have a coauthor graph describing the collaboration
patterns among researchers. Thus, each graph can be viewed as a snapshot
of the overall collaboration network. These graphs can be aggregated in
an OLAP style. For instance, we can aggregate graphs in order to obtain
collaborations by conference type and year for all pairs of authors. For this,
we must aggregate the nodes and edges in each snapshot graph according to
the conference type (like database conferences) and the year. For example, if
there is a link between two authors in the SIGMOD and VLDB conferences,
the nodes and the edge will be in the aggregated graph corresponding to
the conference type Databases . More complex patterns can be obtained,
for example, by merging the authors belonging to the same institution,
enabling to obtain patterns of collaboration between researchers of the same
institutions.
Taking the above concepts into account, in Graph OLAP, dimensions
are classified as informational and topological . The former are close to the
traditional OLAP dimension hierarchies using information of the snapshot
levels, for example, Conference All . They can be used to aggregate

and organize snapshots as explained above. On the other hand, topological
dimensions can be used for operating on nodes and edges within individual
networks. For example, a hierarchy for authors like AuthorId

→ Field →

Institution
will belong to a topological dimension since author institutions do not
define snapshots. These definitions yield two different kinds of Graph OLAP
operations. A roll-up over an informational dimension overlays and joins
snapshots (but does not change the objects), while a roll-up over a topological
dimension merges nodes in a snapshot, modifying its structure.
Graph Cube [ 239 ] is a model for graph data warehouses that supports
OLAP queries on large multidimensional networks, accounting for both
attribute aggregation and structure summarization of the networks. A multi-
dimensional network consists of a collection of vertices, each containing a set
of multidimensional attributes describing the nodes' properties. For example,
in a social network, the nodes can represent persons, and multidimensional
attributes may include UserID , Gender , City , etc. Thus, multidimensional
attributes in the graph vertices define the dimensions of the graph cube.
Measures are aggregated graphs summarized according to some criteria.
Note that the problem here is different from Graph OLAP, where there are
several snapshots. In Graph Cube, we have only one large network, thus we
have a graph summarization problem. For example, suppose that we have
a small social network with three nodes. Two of them correspond to male
individuals in the network, while the third corresponds to a female. A graph
that summarizes the connections between genders will have two nodes, one
labeled male and the other labeled female . The edges between them will be
annotated with the number of connections of some kind. For instance, if in the
original graph there were two connections between two male persons (in both

→