Semantic Approaches for Accessing Distributed Information

XIKE allows the choice of the "severity level" against which the interschema property extraction task is performed. To the best of our knowledge, has not been previously proposed in the literature. It derives from the consideration that applications and scenarios possibly benefiting of interschema properties are numerous and extremely various. In some situations (e.g., in Public Administrations, Finance, and so on) the extraction process must be very severe in that two concepts must be considered equivalent only if their similarity is strengthened by more clues; in such a case a high severity level is required.
In other situations (e.g., tourist Web pages) the extraction task can be looser and can decide to consider equivalent two concepts having some similarities but presenting also some differences.
Specifically, at the beginning of the extraction activity our approach asks the user to specify the desired severity level by means of a suitable, friendly interface; this is the only information required to him until the end of the extraction task, when he has to validate obtained results. Interestingly enough, from the severity level point of view, a classical approach can be seen as a particular case of that presented in this paper in which a severity level is fixed and all concept similarity evaluations are performed against this level.
XIKE is almost automatic. Specifically, it requires the user intervention only in few specific cases. In particular, the only compulsory user intervention is for validating obtained results; this is necessary because the problems we are considering are inherently complex and, therefore, the solutions provided by the machine should be validated and, if necessary, modified by the "intelligence" provided by a human expert.
XIKE has been specialized for operating on XML Schemas; as a consequence it can profitably exploit the specificities of the format it handles. A further contribution of XIKE is its "lightness". As far as this aspect is concerned, two dierent philosophies can be encountered in the literature.
Specifically, some approaches are sophisticated, based on a variety of thresholds, weights, parameters, and so on; they generally obtain very precise results but need quite a heavy tuning phase. On the contrary, other approaches are less complex and try to reduce the exploitation of weights and thresholds; generally, they are less precise than the former ones but do not need a tuning phase; they are particularly suited when involved sources are numerous and large. Our approach belongs to this second category since it does not require
any setting of thresholds or weights, unless the user desires to personalize it; as a consequence, it is simple and light and does not need a tuning activity.
However, in spite of its "lightness", the quality of obtained results is satisfactory.