The challenge is how to implement semantics into BI environments
By Madan Sheina
09 Apr 2009
There’s an increased expectation that business intelligence (BI) and analytic tools will better understand individual user needs and tap into an organisation’s collective intelligence to speed up and enhance corporate decision-making. Can the principles of a semantic web help to practically address the current limitations of BI beyond this and achieve the vision of BI 2.0 – or is this just a new snake oil devised from academic debates? While we believe it’s too early to start to define a real market for semantic BI, the integration of semantics into BI and analytics is an opportunity waiting to happen. If the technical implementation challenges can be overcome then semantic technologies have the potential to advance the use of BI and analytics in interesting and innovative ways.
Semantics has a role to play in realising the BI 2.0 vision
If the BI 2.0 vision is to have any substance to it, then the semantic web might be a big part of it. BI 2.0 implicitly defines a semantic data model that doesn’t require knowledge of data structures, whether it comes from structured relational databases or semi-structured content sources like documents or web pages. It’s the understandability of these models that is one of the core value propositions of BI 2.0, enabling unfettered levels of information access, sharing and collaboration.
However, most BI implementations today lack a way to integrate relevant intelligence found across a broader non-relational domain of data. If BI’s aim is to provide a comprehensive knowledge of all of the factors that affect a business, then advances in semantic web technologies can help, thanks to their ability to distil, understand and link data from a range of heterogeneous structured and unstructured sources.
Most BI systems also fail to provide the exploratory power of ‘inference’ among the data sets they analyse or to take into account the contextual relevancy of user queries. Both are needed to solve the broad and often individualised requests of decision-makers. The original objective of the semantic web is to enable the description of web content using domain-specific data models (called ontologies) that make it possible for applications to locate and reason over the Web’s resources. Similarly, ontologies can feasibly be adapted to model the structure of a data warehouse’s analytically focused star schema.
A BI foundation built on this type of ontological scheme is a powerful proposition since it allows BI queries to make semantic inferences over an extended range of external and distributed data that is not necessarily stored in the data warehouse; rather it is referenced through an ontology. The analysis is also enriched by a deeper semantic understanding of data relationships within the data warehouse itself, and is not restricted to the rigid relationships pre-coded in a star schema.
The challenge is how to implement semantics into BI environments
Semantically driven ontologies might appear to be a superior way to smarten up integrated data sources with richer meaning and context. The big challenge is how to weave them into traditional BI and data warehousing processes and environments.
The Web 2.0 idea of being able to semantically integrate data sourced from across the Web might therefore seem far-fetched given the trials and tribulations of effectively integrating data from different parts of the same company. Moreover, information itself often lacks a meaningful context. The knowledge extracted from unstructured sources must be of sufficient quality that it can be used in BI systems and subjected to analysis in the same way that structured data can be. Finally, building analytically focused ontologies that work with BI queries is a tall order – most SQL and OLAP queries are not easily reproduced in query languages for ontologies.
So far only a few BI vendors have taken a serious stab at addressing these challenges. However, there is some early work being done on semantically driven data integration schemes that attempt to blend and represent data with richer meaning and context. Development work has started on a semantic web equivalent of the SQL query language for accessing relational databases, called Sparql. As development ramps up, prototype applications should start to emerge that practically demonstrate the beneficial links between the semantic web and BI.
