Dynamic Web services composition aims to generate a composition plan at run-time. Semantic-based techniques rely on annotating services to facilitate the discovery of the service components that satisfy a user need (matchmaking). The matchmaking process places most attention on service selection rather than on the behaviour of the composed service, and the service components are arranged considering simple control-flow patterns (mainly sequence). In real life scenarios, however, composed service behaviour follows complex control-flow patterns that satisfy the needs of business processes, which are generally defined through manual service composition. In this paper we present a technique to derive complex composed service behaviour semantics, such semantics make possible to dynamically and automatically discover complex services compositions. We have implemented and tested our technique with a known dataset with better performance when compared to simple service composition strategies.