RELEvant VAlues geNeraTor

Who?

What?

• The idea is that, in many cases, the values of an attribute domain may be clustered because strongly related according to some kind of "hidden relationship".
  
  
• Providing a name to these clusters, we may refer to a relevant value name which encompasses a set of values.
  
  

• More formally, given a class/table C and one of its attributes At, a relevant value for it, rv^At is a pair
  
  rv^At = < rvn^At, values^At >
  where rvn^At is the name of the relevant value, while values^At is the set of values referring to it
  
  

Where?

Publications about Relevant

• S. Bergamaschi, F. Guerra, M. Orsini, C. Sartori, M. Vincini: "RELEVANT News: a semantic news feed aggregator", 4th Workshop on Semantic Web Applications and Perspectives (SWAP 2007), Bari, Italy, December 18-20, 2007
   
• S. Bergamaschi, F. Guerra, M.Orsini, C.Sartori: "Extracting Relevant Attribute Values for Improved Search", IEEE Internet Computing, vol. 11, no. 5, pp. 26-35, Sept/Oct, 2007 (special issue on Semantic-Web-Based Knowledge Management), ISSN 1089-7801
  
  
• S. Bergamaschi, F. Guerra, M. Orsini, C. Sartori: "A new type of metadata for querying data integration systems", Proceedings of the Convegno Nazionale Sistemi di Basi di Dati Evolute (SEBD2007), Torre Canne (Fasano, BR)|, 17-20 June 2007, pp 266-273, ISBN 978-88-902981-0-3
  
  
• S. Bergamaschi, F. Guerra, M. Orsini, C. Sartori: "Relevant values: new metadata to provide insight on attribute values at schema level", In proceedings of the 9th International Conference on Enterprise Information Systems, Funchal, Madeira – Portugal, 12-16, June 2007 (p.p. 274-279). ISBN 978-972-8865-88-7
  
  
• M.Orsini:"An automatic metadata extraction tool for Information Integration Systems ", Poster (pdf) at "D-Day 2007", International Doctorate School in Information and Communication Technologies

When?

Why?

• Our research is motivated form the idea that the knowledge about the metadata describing a database (table's name, attributes' names and domain, ...) is often not enough for writing a query, expecially in a data integration environment like MOMIS.
  
  
  • Integration puts together in the same global class a number of local semantically similar classes coming from different sources.
    
    
  • The name/description of a global class/global attribute is often generic and significantly limiting the effectiveness of querying.
    
    
  • Ignoring the values assumed by a global attribute may generate meaningless, too selective or empty queries.
    
    
  • Knowing all the data collected from a global class is infeasible for a user: databases contain large amount of data which a user cannot deal with.
    
    
• A metadata structure derived from an analysis of the attribute extension could be of great help in overcoming such limitation. Such metadata represent a synthesized extensional knowledge emerging from the attribute values. They are “relevant values” as they provide to the users a synthetic description of the values of the attribute by representing its domain with a reduced number of values.
  
  

How?

1. How can we cluster the values of the domain in order to put together in a relevant value a set of values which are strongly related?
   By means of data mining and clustering techniques, adapted to the problem on hand. The techniques take into account some semantics extracted from:
   1. the syntax of the values: values related to the same object may have the same etymology and then share a common root;
      
   2. the dominance, which discovers values more general than other ones;
      
   3. the lexical meaning w.r.t. WordNet, which identifies semantically related values expressed with a different terminology.
   
   
   
2. How can we choose the relevant value names?
   It will require the intervention of the designer, but we will provide an effective assistant.

Further information

• A running example will be soon available

Last Update: 6 Sep 2007