Service discovery employs matching techniques to select services by comparing their descriptions against user constraints. Semantic-based matching approaches achieve higher recall than syntactic-based ones (as they employ ontological reasoning mechanisms to match syntactically heterogeneous descriptions). However, semantic-based approaches still have problems (e.g. lack of scalability as an exhaustive search is often performed to located services conforming to constraints). This paper proposes two approaches that deal with the problem of scalability/performance for composite service location. First, services are indexed based on the values they assign to their restricted attributes (the attributes restricted by a given constraint). Then, services that assign "conforming values" to those attributes are combined to form composite services. The first proposed approach extends a local optimisation technique to perform the latter, since identifying such values is NP-hard. However, this approach returns false negatives since the local optimisation technique does not consider all the values. Hence, a second approach that derives conforming values using domain rules is defined. The used rules are returned with each composite service so that a user can understand the context in which it is retrieved. Results obtained from the experiments that varied the number of available services demonstrate that the performance of the local optimisation-based approach is 76% better than existing semantic-based approaches and recall is 98% higher than syntactic-based approaches.