Multidimensional reputation network for service composition in the internet of things Lorenzo Bossi DiSTA University of Insubria Italy lorenzo.bossi@uninsubria.it Stefano Braghin IBM Research Dublin Ireland stefanob@ie.ibm.com Alberto Trombetta DiSTA University of Insubria Italy alberto.trombetta@uninsubria.it Abstract—The composition of services, as envisioned in the Internet of Things and Web of Things paradigms, is a most relevant component. To be effective, services composition should choose among the discovered services the most suitable ones and this a non-trivial task given the (typically) very large number of available services. In this work we propose a framework for assessing the reputation of a set of services as a measure for the suitability of such services in coping with a given task. We take as a starting point an approach proposed for team recommendation that aims at maximize the overall team reputation, based on the feedback that each team member receives from other members. It is important to point out that several important differences among the approaches exist – which we highlight in the paper – that do not allow for a mere adaptation from the team recommendation setting. Index Terms—reputation; service composition; Internet of Things; I. I NTRODUCTION Internet Of Things (IoT) and Web of Things (WoT) are two new paradigms aiming at interconnecting devices and objects that contain a (possibly tiny) processing unit (like sensor networks, ambient devices etc.) and to fully integrate them in a web-based fabric. More precisely, the WoT paradigm is based on the idea the interconnection and the integration of such “smart objects” should be obtained by adopting the suite of W3C web standards, from the HTTP protocol to the Semantic Web. On the other hand, the IoT paradigm does not require web standards but works directly on TCP/IP protocols. By taking inputs from heterogeneous sources and exchanging information, an interconnected web of objects may thus provide a set of services, which may in turn compose with other objects/services. Several approaches have been proposed – as we will discuss in Section II – to deal with composition, discovery and inte- gration of services. A relevant issue in this research area is to propose efficient and practical solutions for the identification of the services that offer the best fit for the task to be accomplished in terms of parameters like availability, offered functionalities and trustworthiness – or reputation – of the provider. Services’ reputation is particularly relevant when there is a large number of available services having roughly the same functionalities (and possibly deal with sensitive data). In this case, it is essential to filter out – at least – those having very low reputation (that is, the untrusted ones). In this work we present a framework for IoT services composition which takes into account the reputation of the service provider and the reliability of the reputation provider in order to identify the “best” candidates to the creation of a composed service, to cope with a given task. To address such issue we leverage on a for team formation procedure presented in [1]. According to such approach, the assessment on how “good” a provider is in given service is determined dynamically depending on – among other things – the judgments or feedback ratings that the provider has previously obtained when performing such service, or related ones. Naturally, taking into account providers’ ratings lends itself to a system dealing with the reputations of the providers themselves. That is, providers (by giving ratings about the services offered by other providers and receiving about what they have carried out) form a web of trust such that (i) two providers are directly connected if one rated the output of the other; (ii) the overall reputation of a provider (with respect to a given service) is computed as an aggregate value of the ratings (relative to such service) given by other providers; (iii) the perception of the reputation of provider u from the point of view of another provider v depends on the reputations and ratings of the providers connecting (possibly by multiple hops over the web of trust) providers u and v. An important aspect of the presented work is that we take into account how services may be related among themselves, in terms of their (possible) mutual interchangeability. In other words, we consider the services to be similar (in varying degrees) and this contributes in determining which are the best providers that can cope with the task to be performed. The computation of reputation takes into account many factors that concur in forming the reputation forming the web of trust existing among service providers, e.g. looking at providers’ ratings for each service, along with information about services’ similarities. Note that in this work we are not focusing on mechanisms for service discovery or service specification languages. In what follows will be presented a framework for collectively assessing the reputation of providers that can be easily inte- grated in existing service specification and discovery settings.