The modern environment of mobile, pervasive, evolving ser- vices presents a great challenge to traditional solutions for enabling in- teroperability. Automated solutions appear to be the only way to achieve interoperability with the needed level of flexibility and scalability. While necessary, the techniques used to determine compatibility, as a precursor to interaction, come at a substantial computational cost, especially when checks are performed between systems in unrelated domains. To over- come this, we apply machine learning to extract high-level functionality information through text categorisation of a system's interface descrip- tion. This categorisation allows us to restrict the scope of compatibility checks, giving an overall performance gain when conducting matchmak- ing between systems. We have evaluated our approach on a corpus of web service descriptions, where even with moderate categorisation accuracy, a substantial performance benefit can be found. This in turn improves the applicability of our overall approach for achieving interoperability in the Connect project.