Self-Repairing Semantic Networks
The preceding Chapter presented a model for PDMSs as well as several methods for analyzing syntactic and semantic losses when reformulating a query through a mapping. These methods were used to forward queries in a self-organizing way, and consisted in peers collaborating in order to disseminate a query in a network of loosely-coupled and heterogeneous information parties. In the present Chapter, we take our approach one step further: rather than only guiding searches by the results obtained from the reformulation analyses, advantage is also taken of the network feedback in order to modify the mappings in an automatic manner. Thus, a step is taken towards self-learning networks of peers collaboratively establishing semantic interoperability in an automated fashion [ACMH03a]. Experimental results are provided that demonstrate how the different kinds of semantic analyses of reformulations interact with the modification of potentially unsound mappings. The initial results interpreted below provide promising evidence that Semantic Gossiping can be used to automatically reach semantic agreement in large networks of computer-generated and dynamic mapping links. In particular, they indicate in which cases each of the two semantic similarity measures derived from cycle and result analysis are more suitable, and how our approach scales with different parameters.