I am often asked what the point of the semantic web is and also how people can implement it and be a part of it through their websites. I came across this paper and thought it was an excellent example of the big picture. Wong, Crowder and Shadbolt wrote “Lesson Learnt from a Large-Scale Industrial Semantic Web Application“. It’s interesting because they have a clear real-world problem and successfully used the semantic web model to solve it. Organising vast amounts of data in a useful way and being able to extract knowledge from it at will is magic. I think so. They say that companies like Rolls Royce are aiming to have long-term service maintenance agreements for their fleet. Maintenance cost has to be be lowered for this model to work and to do that the engineers need to gather knowledge from maintenance histories of similar products during the design phase of new products. Different teams are in charge of different parts of the engine so it’s impossible to get through all of this documentation effectively. The information needs to be complete and fresh. This is a problem common in many sectors so it’s interesting to learn how they did this and to think about how it can be applied elsewhere.
Here is the summary:
“The design and maintenance of an aero-engine generates a significant amount of documentation. When designing new engines, engineers must obtain knowledge gained from maintenance of existing engines to identify possible areas of concern. We developed a Semantic Web based document repository for transferring front-line maintenance knowledge to design. The Semantic Web is an ideal candidate for this application because of the size and distributed nature of an aerospace manufacturer’s operation. The Semantic Web allows us to dynamically cross reference documents with the use of an ontology. However, during the design and implementation of this project, we found deficiencies in the W3C1 recommended Semantic Web query language SPARQL. It is difficult to answer questions our users sought from the document repository using SPARQL. The problem is that SPARQL is designed for handling textual queries. In industrial applications, many common textual and semantic questions also contain a numerical element, be it data summarization or arithmetic operations. In this paper, we generalize the problems we found with SPARQL, and extend it to cover web applications in non-aerospace domains. Based on this analysis, we recommend that SQL-styled grouping, aggregation and variable operations be added to SPARQL, as they are necessary for industrial applications of the Semantic Web. At the moment, to answer the non-textual questions we identified with an RDF store, custom written software is needed to process the results returned by SPARQL. We incorporated the suggested numerical functionalities from SQL for an example query, and achieved a 21.7% improvement to the speed of execution. More importantly, we eliminate the need of extra processing in software, and thus make it easier and quicker to develop Semantic Web applications.”
It’s interesting that SPARQL presented so many limitations for them and that they found SQL-style aggregation instead. I’d like to read other instances in industry where the semantic web model has been implemented with or without success so let me know if you can suggest any.
Re. SPARQL you don’t need to go out of SPARQL with custom solution. What you need is an DBMS that combines SPARQL and SQL so that in the absolute worst case you can surface SQL built-in functions as “magic predicates”. That said, you can also use an RDF DBMS that supports Aggregates natively within its SPARQL implementation.
One such example is OpenLink Virtuoso. It supports SPARQL Aggregates (basically SPARQL-BI) and SPASQL (SPARQL intermingled with SQL).
Links:
1. http://delicious.com/kidehen/virtuoso_sparql_tutorial
2. http://delicious.com/kidehen/virtuoso_whitepaper
3. http://delicious.com/kidehen/rdb2rdf
Kingsley
Thanks Kingsley for your input, much appreciated. Checking out those links now.