Paolo Besana

Probabilistic dialogue models for dynamic ontology mapping

Agents need to communicate in order to accomplish tasks that they are unable to perform alone. Communication requires agents to share a common ontology, a strong assumption in open environments where agents from different backgrounds meet briefly, making it impossible to map all the ontologies in advance. An agent, when it receives a message, needs to compare the foreign terms in the message with all the terms in its own local ontology, searching for the most similar one. However, the content of a message may be described using an interaction model: the entities to which the terms refer are correlated with other entities in the interaction, and they may also have prior probabilities determined by earlier, similar interactions. Within the context of an interaction it is possible to predict the set of possible entities a received message may contain, and it is possible to sacrifice recall for efficiency by comparing the foreign terms only with the most probable local ones. This allows a novel form of dynamic ontology matching.

Approximate Structure-Preserving Semantic Matching

Typical ontology matching applications, such as ontology integration, focus on the computation of correspondences holding between the nodes of two graph-like structures, e.g., between concepts in two ontologies. However, there are applications, such as web service integration, where we may need to establish whether full graph structures correspond to one another globally, preserving certain structural properties of the graphs being considered. The goal of this paper is to provide a new matching operation, called structure preserving matching. This operation takes two graph-like structures and produces a set of correspondences between those nodes of the graphs that correspond semantically to one another, (i) still preserving a set of structural properties of the graphs being matched, (ii) only in the case if the graphs are globally similar to one another. We present a novel approximate structure preserving matching approach that implements this operation. It is based on a formal theory of abstraction and on a tree edit distance measure. We have evaluated our solution with encouraging results.

Models of Interaction as a Grounding for Peer to Peer Knowledge Sharing

Most current attempts to achieve reliable knowledge sharing on a large scale have relied on pre-engineering of content and supply services. This, like traditional knowledge engineering, does not by itself scale to large, open, peer to peer systems because the cost of being precise about the absolute semantics of services and their knowledge rises rapidly as more services participate. We describe how to break out of this deadlock by focusing on semantics related to interaction and using this to avoid dependency on a priori semantic agreement; instead making semantic commitments incrementally at run time. Our method is based on interaction models that are mobile in the sense that they may be transferred to other components, this being a mechanism for service composition and for coalition formation. By shifting the emphasis to interaction (the details of which may be hidden from users) we can obtain knowledge sharing of sufficient quality for sustainable communities of practice without the barrier of complex meta-data provision prior to community formation.

How service choreography statistics reduce the ontology mapping problem

In open and distributed environments ontology mapping provides interoperability between interacting actors. However, conventional mapping systems focus on acquiring static information, and on mapping whole ontologies, which is infeasible in open systems. This paper shows that the interactions themselves between the actors can be used to predict mappings, simplifying dynamic ontology mapping. The intuitive idea is that similar interactions follow similar conventions and patterns, which can be analysed. The computed model can be used to suggest the possible mappings for the exchanged messages in new interactions. The suggestions can be evaluate by any standard ontology matcher: if they are accurate, the matchers avoid evaluating mappings unrelated to the interaction. The minimal requirement in order to use this system is that it is possible to describe and identify the interaction sequences: the Open-Knowledge project has produced an implementation that demonstrates this is possible in a fully peer-to-peer environment.

An evaluation of ontology matching in geo-service applications

Matching between concepts describing the meaning of services representing heterogeneous information sources is a key operation in many application domains, including web service coordination, data integration, peer-to-peer information sharing, query answering, and so on. In this paper we present an evaluation of an ontology matching approach, specifically of structure-preserving semantic matching (SPSM) solution. In particular, we discuss the SPSM approach used to reduce the semantic heterogeneity problem among geo web services and we evaluate the SPSM solution on real world GIS ESRI ArcWeb services. The first experiment included matching of original web service method signatures to synthetically alterated ones. In the second experiment we compared a manual classification of our dataset to the automatic (unsupervised) classification produced by SPSM. The evaluation results demonstrate robustness and good performance of the SPSM approach on a large (ca. 700 000) number of matching tasks.

The benefits of service choreography for data-intensive computing

As the number of services and the size of data involved in workflows increases, centralised orchestration techniques are reaching the limits of scalability. In the classic orchestration model, all data pass through a centralised engine, which results in unnecessary data transfer, wasted bandwidth and the engine to become a bottleneck to the execution of a workflow. Choreography techniques, although more complex to model offer a decentralised alternative and are the optimal architecture for data-centric workflows; data are passed directly to where they are required, at the next service in the workflow. While orchestration is the dominant architectural approach, there are relatively few choreography languages and even fewer concrete implementations. This papers contributions are twofold. Firstly we argue the case for choreography in data-intensive computing, and demonstrate through workflow patterns the advantages in terms of scalability when a choreography architecture is adopted. Secondly we introduce the Light Weight Coordination Calculus (LCC), a type of process calculus used to formally define choreographies, and the OpenKnowledge framework, a choreography-based architecture, providing the functionality for peers to coordinate in an open peer-to-peer system. Through LCC and the OpenKnowledge framework we practically demonstrate how choreography can be achieved in a lightweight manner with a comparatively simple process language.

An interaction-centric approach to support peer coordination in distributed emergency response management

Modern information systems are required to operate in distributed and dynamic environments. In such open settings, coordination technologies play a crucial role in the design of flexible software systems. Research efforts in different areas are converging to devise suitable mechanisms for process and peer coordination: in particular, current results on service-oriented computing and multi-agent systems are being integrated to support dynamic decision-making processes among autonomous components in large, open systems. This paper addresses how agent technologies can be designed, applied, and eventually integrated with standard technologies, in order to build more robust and intelligent systems. The focus of our research is on the engineering, exploitation and evaluation of an agent protocol language in realistic contexts. In particular, a specific executable protocol language is adopted to specify simulated interactions among distributed processes which will be tested in emergency response domain activities (that we will refer to hereafter as e-Response activities), chosen as an example of knowledge-intensive and dynamic application domain where intelligent decision making is crucial. We present a novel approach based on shared protocols models distributed through a peer-to-peer infrastructure and we show how it can be applied in the context of crisis management to support coalition formation and process coordination in open environments. Specifically, a prototype e-Response simulation system - built on a Peer-to-Peer (P2P) infrastructure - has been developed to execute interaction models describing common coordination tasks in the emergency response domain. Preliminary evaluation of the proposed framework demonstrates its capability to support such e-Response tasks.

An Executable Calculus for Service Choreography

The Lightweight Coordination Calculus (LCC) is a compact choreography language based on process calculus. LCC is a directly executable specification and can therefore be dynamically distributed to a group of peers for enactment at run-time; this offers flexibility and allows peers to coordinate in open systems without prior knowledge of an interaction. This paper contributes to the body of choreography research by proposing two extensions to LCC covering parallel composition and choreography abstraction. These language extensions are evaluated against a subset of the Service Interaction Patterns, a benchmark in the process modelling community.

OpenKnowledge for peer-to-peer experimentation in protein identification by MS/MS.

BackgroundTraditional scientific workflow platforms usually run individual experiments with little evaluation and analysis of performance as required by automated experimentation in which scientists are being allowed to access numerous applicable workflows rather than being committed to a single one. Experimental protocols and data under a peer-to-peer environment could potentially be shared freely without any single point of authority to dictate how experiments should be run. In such environment it is necessary to have mechanisms by which each individual scientist (peer) can assess, locally, how he or she wants to be involved with others in experiments. This study aims to implement and demonstrate simple peer ranking under the OpenKnowledge peer-to-peer infrastructure by both simulated and real-world bioinformatics experiments involving multi-agent interactions.MethodsA simulated experiment environment with a peer ranking capability was specified by the Lightweight Coordination Calculus (LCC) and automatically executed under the OpenKnowledge infrastructure. The peers such as MS/MS protein identification services (including web-enabled and independent programs) were made accessible as OpenKnowledge Components (OKCs) for automated execution as peers in the experiments. The performance of the peers in these automated experiments was monitored and evaluated by simple peer ranking algorithms.ResultsPeer ranking experiments with simulated peers exhibited characteristic behaviours, e.g., power law effect (a few dominant peers dominate), similar to that observed in the traditional Web. Real-world experiments were run using an interaction model in LCC involving two different types of MS/MS protein identification peers, viz., peptide fragment fingerprinting (PFF) and de novo sequencing with another peer ranking algorithm simply based on counting the successful and failed runs. This study demonstrated a novel integration and useful evaluation of specific proteomic peers and found MASCOT to be a dominant peer as judged by peer ranking.ConclusionThe simulated and real-world experiments in the present study demonstrated that the OpenKnowledge infrastructure with peer ranking capability can serve as an evaluative environment for automated experimentation.

Distributed Workflows: The OpenKnowledge Experience

Software systems are becoming ever more complex, and one source of complexity lies in integrating heterogeneous subsystems. Service Oriented Architectures are part of the answer: they decouple the components of the system. However normally SOA is used from a centralised perspective: a single process invokes remote services, unaware of being part of a workflow. We claim that the centralised, or orchestration-based, approach cannot scale well with increasing complexity and heterogeneity of the components, and we propose an alternative distributed, or choreography-based, approach, that forces developers to think in terms of actors, roles and interactions. We first present the OpenKnowledge framework, designed according to choreography-based principles and then show how a complex, distributed model for managing the triple assessment of patients suspected with breast cancer can be easily implemented using this framework.