Traditional service composition approaches rely mostly on centralised architectures, which have been proven inadequate in pervasive Internet of Things (IoT) environments. In such settings, where decentralisation of decision-making is mandatory, nature-inspired computing paradigms have emerged due to their inherent capability to accommodate spatiality, self-adaptivity, and evolvability. In this paper, taking inspiration from natural metaphors we propose a decentralised service composition model which is based on artificial potential fields. In the proposed approach, artificial potential fields (APFs) lead the service composition process through the balance of forces applied between service requests and service nodes. APFs are formed considering the percentage of user requested services that can be offered by service provision nodes, as well as service node availability. Τhe applicability of the proposed approach is discussed in an exemplar scenario concerning dynamic and personalised composition of an audio-visual virtual guide service in an IoT network of a trade show venue.