Information systems of every organization (ranging from large companies to individual entities) have to handle a variety of information sources, from proprietary ones to information publicly available in web services worldwide. Grasping relevant information wherever it may be and exchanging information with all potential partners has become an essential challenge. Basically, information sharing, rather than information processing, is IT's primary goal in the 21st century. The key point is that now information has to be sharable in an open environment, where interacting peers do not necessarily have a common understanding of the world at hand, as used to be the case in traditional enterprise information systems. Lack of common background generates the need for explicit guidance in understanding the exact meaning of the data, i.e., their semantics. Data semantics is more and more context- and time-dependent, and cannot be xed once and for all at design time. Identifying emerging relationships among previously unrelated information items (e.g., during data interchange) may dramatically increase their value. Such relationships are the basic ingredients for semantic interoperability that is viewed as an emergent phenomenon constructed incrementally, and its state at any given point in time depends on the frequency, the quality and the e ciency with which negotiations can be conducted to reach agreements on common interpretations within the context of a given task. This type of semantic interoperability is referred to as \emergent semantics" [3, 5].
Software agents have various mechanisms at their disposal
for establishing relationships between internal symbols and
external meaning. In many cases, humans are responsible
for providing the initial semantics. In the simplest case,
the natural language vocabulary is used for the local
symbols and their relationship with the de nition of the notion
concerned is left implicit. Often, the hidden assumption
is that the local symbol meaning is identi ed through
human cognition. In order to address some of the problems
arising when leaving interpretation of the symbol implicit
semantics to human cognition, some researchers have
proposed to use an explicit, shared reference system for relating
sets of symbols. Ontologies serve this purpose: they consist
of explicit, partial de nitions of the intended meaning of
symbols for a domain of discourse. Unfortunately,
building shared ontologies is a complex process and top-down
ontology design, even when done collaboratively, is known
not to scale well. Moreover, ontologies are not enough to
achieve semantic interoperability. For instance, ontologies
are forms of \a-priori" agreements on concepts, and
therefore, their use is insu cient in ad-hoc and dynamic
situations where the interacting parties did not anticipate all the
interpretations and where \on-the- y" integration must be
performed [
An example of systems dealing with semantic
interoperability in dynamic open environments, i.e. emergent semantics,
is Esteem (Emergent Semantics and cooperaTion in
multiknowledgE EnvironMents) [
The Esteem approach proposes a comprehensive framework and platform for data and service discovery in P2P systems, with advanced solutions for trust and quality-based data management, P2P infrastructure de nition, query processing and dynamic service discovery in a context-aware scenario. The system allows one to access data and services in a simple and e ective way, by querying information sources that are similar to the user's interests. Common interests identify semantic communities, which represent semantic a nity between peers emerging in a dynamic and heterogeneous environment. Data and service discovery is performed inside the borders of such communities.
Semantic matchmaking Network & overlay
Context Matching Service Matching Ontology Matching Context Manager Quality Manager
Query
Manager P2P Mapping
Manager Community Membership
Semantic Neighbor
Semantic
Routing Global Overlay
Semantic Overlay
Preferential
Link
Data & service discovery Semantic community & routing Semantic communities are created and updated in an automatic way by collecting information sources whose contents present high similarity. A threshold-based mechanism allows to establish internal cohesion of community contents, enabling peer aggregation apart of their terminological differences. Semantic communities do not constrain participants to adhere to a global ontology, but compare reference ontologies used by information sources belonging to the same community. In the Esteem system context-aware data and service selection excludes from search results resources that are not accessible from the particular user context. The Esteem system is also in charge of protecting the users from retrieving data and services from untrustworthy information sources.
Esteem relies on an overlay P2P network where i) semantic communities are de ned to aggregate peers with similar interests and ii) a probe/search mechanism is adopted to enforce data and service discovery/sharing. An Esteem semantic community sc is de ned as a pair of the form sc = hCID; M i where CID is the unique Community Identier that characterizes the community sc and M is the Manifesto, that is the community ontology that describes the common interpretation (i.e., perspective) of the community interests. In Esteem, a semantic community is autonomously emerging, in that it originates from a proposal of a community founder (i.e., a peer) which initiates the community formation through dissemination of an advertisement message that contains CID and M of the emerging community. Each receiving peer pi autonomously decides whether to join the community on the basis of its level of interest in the received manifesto M . Such a level of interest is computed by invoking an ontology-based semantic matchmaker and by evaluating the semantic a nity between M and the peer ontology of pi. Furthermore, communities are exploited as a semantic overlay on top of the basic P2P overlay (i.e., the global overlay) in order to enforce e ective data and service sharing. In this respect, the probe/search mechanism is used to characterize: the discovery phase, based on ontology matching, where probe queries are de ned to identify the peers that are capable of providing relevant knowledge with respect to a given topic of interest; the sharing phase, based on P2P mapping de nition, where standard search queries are issued to point-topoint interact with a previously discovered peer for actual data acquisition and/or service invocation. In the discovery phase, the joined semantic communities are exploited by a requesting peer for selecting the probe query recipients with the aim of choosing those communities and peers that are most likely to provide relevant results according to the query target. In this context, the semantic matchmaker is invoked to evaluate the relevance of a community with respect to a probe query by comparing the community manifesto against the query content. By collecting probe query replies, a peer evaluates the results and decides whether to perform the sharing phase by directly interacting with the most interesting peers that provided a reply through appropriate search queries with the aim of accessing their data and services.
An Esteem peer is equipped with (all or a subset of): (i) a semantic description of shared data and services, to properly identify its interests, expressed through ontologies; (ii) the representation of context(s) from which the peer accesses data and invokes services; (iii) the representation of quality and trust metadata attached to its data and services (quality pro le). When joining semantic communities that share its interests, the peer also maintains information about joined communities. The Esteem architecture addresses the main requirements of a peer involved in P2P semantic cooperation. As shown in Figure 1, the main components are: Network & overlay. It is responsible for managing the peer connectivity and for handling incoming and outgoing messages. From the network point of view, the Esteem P2P infrastructure is organized in semantic overlays featuring the semantic communities. In this respect, the network & overlay component is responsible for maintaining the overlays and the associated peer communications.
Semantic community & routing. It is responsible for managing the peer participation in semantic communities and for discovering the semantic neighborhoods of a peer. Furthermore, this component is responsible for providing a semantic routing mechanism to e ectively enforce query propagation.
Semantic matchmaking. It is responsible for providing semantic a nity evaluation when comparing di erent peers' ontological descriptions. This component is invoked by a peer during the discovery phase to identify peers that are capable of providing matching resources (i.e., data, service, context) w.r.t. a given target request. Di erent techniques are provided by the semantic matchmaking component according to the type of matching resource that is speci ed in the request. In particular, ontology, service, and context matching techniques are provided by the semantic matchmaker. Data & service discovery. It is responsible for interacting with the user and for satisfying its discovery requests. In particular, this component provides the functionalities for context and quality/trust management. Furthermore, discovery and sharing functionalities are also addressed in this component through query/answer and P2P mapping management.
Moreover, the system supports the (human) user with an
intuitive Web interface that assists him/her in joining the
semantic communities that share his/her interests and in
identifying data and services he/she is looking for. The Esteem
system has been validated with a set of specialists from the
health-care domain both to collect system requirements in
the rst stage of the project and to test the system through
a think aloud evaluation technique [