Conclusion - Data Warehouse Systems: Design and Implementation - page 575

There are many open issues to solve in order to make music warehouses
a reality. Further, there is still no clear data model and query language.
Finally, the authors identify ten challenges for music warehouses. Some of
them are the definition of appropriate aggregation functions for acoustic
data, precision-aware retrieval, support of various kinds of hierarchies, and
integration of new data types. As it can be seen, most of the problems already
solved in traditional data warehouses are still open in the music setting, thus
opening an interesting research field for the years to come.
15.5 Graph Analytics and Graph Data Warehouses
Graph analytics has been steadily gaining momentum in the data man-
agement community in recent years since many real-world applications
are naturally modeled as graphs, in particular in the domain of social
network analysis. A graph database management system [ 178 ] is a database
management system that allows creating, reading, updating, and deleting a
graph data model. Some systems use native graph storage, which means
they are optimized and designed for storing and managing graph data
structures. Others serialize the graph data into a relational or an object-
relational database. Graph databases provide better support for graph data
management than relational databases. This is mainly due to the fact that
relational databases deal just implicitly with connected data, while graph
databases store actual graphs. Representative graph databases like Neo4J 2
and Titan 3 have their own data model. They also have their own query
language, called Cypher and Gremlin, respectively.
Given the extensive use of graphs to represent practical problems,
multidimensional analysis of graph data and graph data warehouses
are promising fields for research and application development. There is a
need to perform graph analysis from different perspectives and at multiple
granularities. This poses new challenges to the traditional OLAP technology.
Graphs whose nodes are of the same kind are referred to as homogeneous .
Heterogeneous graphs, on the other hand, can have nodes of different kinds.
We next comment on some proposals based on homogeneous graphs since
work on heterogeneous graphs (e.g., [ 235 ]) is at a preliminary stage.
A first framework and classification of OLAP for graphs was proposed
in [ 28 , 29 ]. This framework, called Graph OLAP , presents a multidimensional
and multilevel view of graphs. As an example, consider a set of authors
working in a given field, say data warehouses. If two persons coauthored
x papers in a conference, say DaWaK 2009, then a link is added between
2 http://www.neo4j.org/
3 http://thinkaurelius.github.io/titan/