Initial symbolic data mining or machine learning methods have been dealing with objects described by sets of attributes, therefore neglecting the substantial information carried by the known relations between objects.
More recently, new methods have been proposed to take into account object relationships to some extent. Within relational data mining, the graph-based data mining methods are the most computationally intensive ones, since they search for explicit graph patterns in data. This type of approaches is particularly well-suited to problems where topological information induced by object relations plays a crucial and subtle role, so that accuracy in the results may balance the high computational cost of these methods. A typical example is organic chemistry where covalent bonds linking atoms represent fundamental pieces of information to explain properties of molecules and chemical reactions. More generally, one can always model any kind of symbolic data (either semi-structured documents in XML, Web Semantic documents in RDF or relational databases) as labelled graphs and benefit from graph mining methods to improve knowledge about data.