Catálogo de publicaciones - libros

Effienct and reliable integration of web data requires building programs called wrappers. Hand writting wrappers is tedious and error prone. Constant changes in the web, also implies that wrappers need to be constantly refactored. Machine learning has proven to be useful, but current techniques are either limited in expressivity, require non-intuitive user interaction or do not allow for -ary extraction. We study using tree-patterns as an extraction language and propose an algorithm learning such queries. It calculates the most information-conservative tree-pattern which is a generalization of two input trees. A notable aspect is that the approach allows to learn queries containing both child and descendant relationships between nodes. More importantly, the proposed approach does not require any labeling other than the data which the user effectively wants to extract. The experiments reported show the effectiveness of the approach.

A key issue in open multiagent systems is that of solving the difficulty of an agent to understand messages coming from other agents having different ontologies. is a new way of facing this issue, by exploiting techniques that allow the agents of a MAS to reach mutually acceptable agreements on the exchanged terms. The produced scenario is similar to that of human discussions, where human beings try to solve those situations in which the involved terms are not mutually understandable, by the semantics of these terms. The HISENE approach is a recent JADE-based protocol effectively supporting semantic negotiation. It is based on the idea that an agent that does not understand a term can automatically require the help of other agents that it considers particularly reliable. However, HISENE does not take in account either the possibility of wrong answers coming from the requested agents or the fact that a term can have different meanings. In order to cover these two important issues, in this paper we present an extension of HISENE, called HISENE2, and we show experimentally that it performs better than HISENE with respect both to the quality and the efficiency of the semantic negotiation.

In this paper we present a multi-agent system aiming at supporting patients to search health care services of their interest in an e-health scenario. The proposed system is HL7-aware in that it represents both patient and service information according to the directives of HL7, the information management standard adopted in medical context. In this paper we illustrate the technical characteristics of our system and we present a comparison between it and other related systems already proposed in the literature.

In this paper, we present a web-based, machine-learning enhanced news reader (PersoNews). The main advantages of PersoNews are the aggregation of many different news sources, machine learning filtering offering personalization not only per user but also for every feed a user is subscribed to, and finally the ability for every user to watch a more abstracted topic of interest by employing a simple form of semantic filtering through a taxonomy of topics.

The use of ontologies in the fields of information retrieval and semantic web is well-known. Since long time researcher are trying to find ontological representations of the diverse laws to have a mechanism to retrieve fine granular legal information about diverse legal cases. However, one of the common problems software systems are faced with in constitutional states is the adapting of the diverse privacy directives. This is a very complex task due to lacks in current software solutions – especially from the architectural point of view. In fact, we miss software solutions that manage privacy directives in a central instance in a structured manner. Even more, such a solution should provide a fine granular access control mechanism on the data entities to ensure that every aspect of the privacy directives can be reflected. Moreover, the whole system should be transparent, comprehensible, and modifiable at runtime. This paper provides a novel solution for this by means of ontologies. The usage of ontologies in our approach differs from the conventional form in focusing on generating access control policies which are adapted from our software framework to provide fine granular access on the diverse data sources.

Robust and scalable user context management is the key enabler for the emerging context- and situation-aware applications, and ontology-based approaches have shown their usefulness for capturing especially context information on a high level of abstraction. But so far the problem has not been approached as a data management problem, which is key to scalability and robustness. The specific challenges lie in the imperfection of high-level context information, its time-dependence and the variability in the dynamics between its different elements. The approach presented in this paper presents a layered data model which structures the problems and is geared towards flexible and efficient query processing in combination of relational database and logic-based techniques. The techniques have been successfully applied for context-aware corporate learning support.

This paper outlines the first experiences of an approach for automatically deriving information demands in order to provide users with demand-driven information supply and decision support. The presented approach is based on the idea that information demands with respect to work activities can be identified by examining the contexts in which they exist and that a suitable source for such contexts are Enterprise Models. However, deriving contexts manually from large and complex models is very time consuming and it is therefore proposed that a better approach is to, based on an Enterprise Model, produce a domain ontology and from this then automatically derive the information demand contexts that exist in the model.

The paper proposes a framework to assess the semantic similarity among instances within an ontology. It aims to define a sensitive measurement of semantic similarity, which takes into account different hints hidden in the ontology definition and explicitly considers the application context. The similarity measurement is computed by combining and extending existing similarity measures and tailoring them according to the criteria induced by the context. Experiments and evaluation of the similarity assessment are provided.

Several authors have suggested similarity measures for objects labeled with terms from a hierarchical taxonomy. We generalize this idea with a definition of information-theoretic similarity for taxonomies that are structured as directed acyclic graphs from which multiple terms may be used to describe an object. We discuss how our definition should be adapted in the presence of ambiguity, and introduce new similarity measures based on our definitions.

We present an implementation of our measures that is integrated with a relational database and scales to large taxonomies and datasets. We evaluate our measures by applying them to an object-matching problem from bioinformatics, and show that, for this task, our new measures outperform those reported in the literature. We also verified the scalability of our approach by applying it to patent similarity search, using patents classified with terms from the taxonomy defined by the United States Patent and Trademark Office.

The Semantic Web needs methodologies to accomplish actual commitment on shared ontologies among different actors in play. In this paper, we propose a machine learning approach to solve this issue relying on classified instance exchange and inductive reasoning. This approach is based on the idea that, whenever two (or more) software entities need to align their ontologies (which amounts, from the point of view of each entity, to add one or more new concept definitions to its own ontology), it is possible to learn the new concept definitions starting from shared individuals (i.e. individuals already described in terms of both ontologies, for which the entities have statements about classes and related properties); these individuals, arranged in two sets of positive and negative examples for the target definition, are used to solve a learning problem which as solution gives the definition of the target concept in terms of the ontology used for the learning process. The method has been applied in a preliminary prototype for a small multi-agent scenario (where the two entities cited before are instantiated as two software agents). Following the prototype presentation, we report on the experimental results we obtained and then draw some conclusions.