In this section we briefly discuss how we compare the semantic specification of user query profiles and the semantic specification of source content and capability profiles, and conduct two-levels of reasoning about the semantic relevance between a user query and a collection of data sources.
In DIOM, the semantic relevance reasoning is mainly conducted in the query routing stage, the first step in the DIOM dynamic query processing framework. The detailed routing strategies and algorithms are beyond the subject of this paper. Readers who are interested in further details may refer to [5].
For level-one relevance reasoning, we compare the scope specification of a user query Q with the content and category specification of the set of available data sources. This step serves as a first-round selection which rules out all the information sources that are irrelevant in terms of the query scope description and the source content descriptions. For our example query, level-one relevance reasoning will prune the information sources whose contents are not relevant to books or book reviews. Using the list of information sources in Figure 3, among many others, it will find that Source 7 is not relevant to the query answer. The main idea of level-one relevance reasoning is to prune out those data sources that have no source relations matching one of the interface names in the query scope description of Q.
For level-two relevance reasoning, we compare the capacity specification of a query Q with the query capability descriptions of data sources. During the level-two semantic relevance reasoning, we prune the data sources that have level-one relevance but do not offer enough query capability to contribute to the answer of Q, either due to the restriction on the scope of query parameters of Q, or the restriction on the list of input or output arguments of the sources, or due to the conflict of query interest with the access constraints associated with the sources. For our example query, level-two relevance pruning will further prune away Sources 1, 3, 5, because they are not capable of contributing to the query answer due to the restrictions on either source access or source category. The main idea of level-two relevance reasoning is to prune those data sources (1) that have no input arguments, which are relevant to the input arguments used in the user query, or (2) that have no output arguments, which are relevant to the output arguments used in the user query, or (3) that have conflict with the interest of the user query (such as the query selection conditions do not match the access constraints of the sources).