Hypermedia Multi-Agent Systems (HMAS) promote an alignment between MAS engineering and the Web architecture to enable development of large, open, dynamic and long-lived interaction systems. HMAS environments exploit HATEOAS to facilitate the discovery of the resources whose afordances are required by agents. An insuficiently addressed issue in current HMAS frameworks is that of enabling authorized access to resources in a manner exploiting the dynamics of multi-agent environments. We propose a framework for context-based authorizations for access and discovery of resources in an HMAS, inspired by work on Attributed-Based Access Control and RDF Stream Reasoning. We detail the design of the framework functionality and the integration with current HMAS platforms, highlighting advantages and challenges of the approach.
CEUR ceur-ws.org
Hypermedia Multi Agent Systems (HMAS) are a design paradigm promoting MAS engineering
that is aligned with the web architecture to enable large, open, dynamic and long-lived
interaction systems [
1, such that artifacts, organized into workspaces are represented on the web as resources having properties and afording
actions and event notifications.
HMAS are characterized by an increased interaction dynamics, since agents can enter or exit an environment, can opt to change the role they play in a situation, can start or stop providing a service. This increased logical mobility of agents and the open and long-lived nature of HMAS create the need to apply context-aware boundaries to the interactions that take place. In turn, this requires a dynamic control of authorized access to the services provided by artifact resources in a hypermedia MAS environment.
A simple scenario exemplifies the matter: a research facility implementing a visual event notification system using smart lights in each lab. The smart lights are available as artifacts CEUR Workshop Proceedings in an HMAS environment and digital assistants of researchers can discover and use them. However, management requires that interaction with any smart light be allowed only for employed personnel who are physically present in the labs.
The dynamic access control issue remains insuficiently addressed (see preliminary design
work on signifiers [
The CASHMERE context-based authorization mechanism (illustrated in Figure 1) exploits an opportunity cost of combining the following semantic web technologies.
Context Representation (Step 1 in Fig. 1). We interpret context as ”any information that
can be used to characterize the situation of an entity. An entity is a person, place, or object that
is considered relevant to the interaction between a user and an application, including the user
and applications themselves” [
Shared Context Modeling (Step 2 in Fig. 1). The CONSERT Model identifies
ContextDimensions and ContextDomains as means to manage and provision context information in a
system [
context within an application domain if a subset of context information from or about them is included in at least one ContextDomain of the application. The shared context relation is formalized as a group membership whereby entities that share context belong to the same ContextDomain Group.
RDF Stream Processing for Shared Context Identification (Step 2 in Fig. 1). Context
information can change in time (e.g. due to agent mobility, changes in current activity) and this
requires reasoning methods which can manage such changes. We identify conditions for shared
context by using RDF Stream processing (RSP) frameworks. Specifically, we use the newest
C-SPARQL [
SOLID Web Access Control Authorizations (Step 3 in Fig. 1). CASHMERE uses WAC specifications and the ACL ontology to transform the identified shared ContextDomain membership into an actionable authorization mechanism. Authorized agents are identified by a WebID 3. An artifact in an HMAS advertises its list of authorizations using a rel=acl Link header. Authorizations provide Read, Write, Append or Control access. The acl:agentGroup mode of identifying authorization subject is used in CASHMERE, describing an instance of a vcard:Group which can contain individual agents as members. This maps directly to the mechanism of ContextDomain membership CONSTRUCT outputs.
The novelty of our work lies in reasoning about context information itself as the main conditioning factor to grant authorizations, leading to a very flexible and extensible method of defining situations that count as an interest and legitimate ground for a human or software agent to access data and services in a hypermedia environment.
Figure 2 shows the extension of element types (Artifacts, Workspaces, Environments, Search
Services) of the Yggdrasil HMAS platform [
authorization determination mechanism. Workspaces enable inclusive access (any agent having a FOAF profile) by default (item B. in Fig 2), while individual artifacts dynamically restrict access based on ContextDomain membership (item A. in Fig 2). Control for change in access control policies is defined at HMAS environment level (item C. in Fig 2). The HMAS platform facilitates the validation of a context-based authorization for all HMAS resources that use a context-based access. It does so by running a federated SPARQL query against all ContextDomain Groups observed by a HMAS resource).
CASHMERE exploits a technological fit to ofer a modular solution for context-based access authorization in A&A HMAS platforms, using a flexible context-representation method (CONSERT), a time-aware shared context identification method (RSP using C-SPARQL) and an authorization representation and validation mechanism following SOLID WAC specifications.
This work has been supported by funding under grant agreement PN-III-P1-1.1-PD-2021-0756 from the Romanian National Research, Development and Innovation Plan.