Content-based publish/subscribe offers a convenient abstraction for information producers and consumers, supporting a large-scale system design and evolution by integrating several distributed independent application systems. Unlike in the traditional address-based unicast or multicast, its core problem is how to match events by predicates on the content of events. In existing matching approaches, matching predicates are composed by the conjunction and disjunction of nonsemantic constraints. But, in context of enterprise application integration, although they can match events by their contents, this traditional matching predicates are not expressive enough in manipulating the complex event matching, such as the "one-tomany" and "many-to-one" matching. Therefore, traditional matching approaches should be extended to solve the complex matching problems. After analyzing information matching patterns in enterprise application integration, we propose three matching models, extend this simple matching approach to the multisemantic matching approach and further introduce the temporal constraint variable. The multi-semantic matching approach allows using different operations in accordance with different semantics; the temporal constraint variable supports processing several discrete events in temporal sequences. Then, we extend OBDD graphs into hierarchy coloured OBDD graphs and prove the equivalence of the transformation. Based on the extended OBDD graphs, the composite matching algorithm is presented and analysed. By experiments, we show the proposed algorithm is efficient.