[Developers] Constructing MQL queries automatically

Vanessa Lopez v.lopez at open.ac.uk
Mon Nov 10 13:36:19 UTC 2008


Hi all,

Thank you very much for your answers,  We have already developed a  
Question Answering system over the Semantic Web. The user can type a  
NL query and the system looks for answers found in any of the multiple  
ontologies and RDF documents offered by the Semantic Web (i.e. if  
somebody asks for the members of the rock group nirvana, the system is  
able to find ontologies about music that contain the musicians that  
play in the group Nirvana, which genre is rock). For that we use both   
a semantic gateway calls Watson to automatically crawl data from the  
Web  (similar to swoogle), and our own repositories that we manually  
collect and put online in a Sesame server. The system is call  
PowerAqua:  http://kmi.open.ac.uk/technologies/poweraqua/

However, most of the queries can not be answered, not because the  
system fails to map the user query into 1 or many ontologies, but  
because the ontologies are not populated, contains no answer, or not  
good data (for instance, in an study we did one year ago, we realized  
that the ontologies out there in the Semantic Web cover less than 20%  
than the topics used on evaluations like TREC)  ...  Freebase seems  
like a  dream to me, a lot of good data in a structured format  ...   
With Freebase we can answer millions of queries!!!.    However the  
data is a bit different to the one I use originally (I am using RDF  
and OWL) ... So, I am studying the best way to query freebase ...

Before trying to create my own set of Freebase types as suggested by  
Shawn I'll try to upload the freebase data dumps into Sesame n- 
triples, it may be easier to query than using NQL .. The reason for  
this is that "openness" is the main assumption of our system, in the  
sense  that we do not know in advance which ontology will answer the  
query , and we do not make any assumptions about the sources. In other  
words, we do not make or create explicit rules about the data,  the  
mappings should be on the fly, and if possible in real time ...  But I  
wanted to explore all possibilities first, I saw there are already  
other QA systems over Freebase, like Powerset, so although they use  
different approach than us, I though someone may have come to a better  
solution by using NQL ...  (A better solution than exploring all types  
and how they link, i.e.:
"type" : "/type/type"
  "name" : "film"
)

Thanks again! any suggestion is very useful for me.

Vanessa,

Research Fellow at Knowledge Media Institute
The Open University.
United Kingdom 


More information about the Developers mailing list