Catálogo de publicaciones - libros

Workflow technology has delivered effectively for a large class of business processes, providing the requisite control and monitoring functions. At the same time, this technology has been the target of much criticism due to its limited ability to cope with dynamically changing business conditions which require business processes to be adapted frequently, and/or its limited ability to model business processes which cannot be entirely predefined. Requirements indicate the need for generic solutions where a balance between process control and flexibility may be achieved. In this paper we present a framework that allows the workflow to execute on the basis of a partially specified model where the full specification of the model is made at runtime, and may be unique to each instance. This framework is based on the notion of process constraints. Where as process constraints may be specified for any aspect of the workflow, such as structural, temporal, etc. our focus in this paper is on a constraint which allows dynamic selection of activities for inclusion in a given instance. We call these cardinality constraints, and this paper will discuss their specification and validation requirements.

Rule-based Intrusion Detection Systems (IDS) rely on a set of rules to discover attacks in network traffic. Such rules are usually hand-coded by a security administrator and statically detect one or few attack types: minor modifications of an attack may result in detection failures. For that reason, signature based classification is not the best technique to detect novel or slightly modified attacks. In this paper we approach this problem by extracting a set of features from network traffic and computing rules which are able to classify such traffic. Such techniques are usually employed in off line analysis, as they are very slow and resource-consuming. We want to assess the feasibility of a detection technique which combines the use of a common signature-based intrusion detection system and the deployment of a data mining technique. We will introduce the problem, describe the developed architecture and show some experimental results to demonstrate the usability of such a system.

With the introduction of clustered messaging brokers and the fault-tolerant Mobile Connector, we can guarantee the exactly-once consumption of messages by agents. The context-aware messaging allowed us to decrease the messaging overhead which has to be present in any fault-tolerant solution. This paper proposes a complete fault-tolerant layer for multi-agent systems (EFTL) that does not restrict agent autonomy and mobility in any way. An application can choose if it wants EFTL support and that decision is based on support costs. A persistent publish/subscribe messaging model allows the creation of an external platformindependent fault-tolerant layer. In order to support the multi-agent platforms of different vendors, a large part of the application logic is moved from those platforms to an application server. We present the EFTL system architecture, the algorithm of exactly-once message consumption and the system’s performance analysis.

Most search engines do their text query and retrieval based on keyword phrases. However, publishers cannot anticipate all possible ways in which users search for the items in their documents. In fact, many times, there may be no between a search phrase and descriptions of items that are perfect “hits” for the search. We present a highly automated solution to the problem of between item information and search phrases. Our system can learn rule-based definitions that can be ascribed to with dynamic connotations by extracting structured item information from product catalogs and by utilizing a frequent itemset mining algorithm. We present experimental results for a realistic e-commerce domain. Also, we compare our rule-mining approach to vector-based relevance feedback retrieval techniques and show that our system yields definitions that are easier to validate and perform better.

In this work we present a decoupled architectural approach that allows Software Agents to interoperate with enterprise systems using Web Services. The solution leverages existing technologies and standards in order to reduce the time-to-market and increase the adoption of agent-based applications. Insights on applications that may be enhanced by the model are presented.

Documents are ubiquitous in modern desktop environments, yet these environments are based on the notion of application rather than document. As a result, editing a document often requires juggling with several applications to edit its different parts. This paper presents OpenDPI, an experimental user-interface toolkit designed to create document-centered environments, therefore getting rid of the concept of application. OpenDPI relies on the DPI (Document, Presentation, Instrument) model: documents are visualized through one or more presentations, and manipulated with interaction instruments. The implementation is based on a component model that cleanly separates documents from their presentations and from the instruments that edit them. OpenDPI supports advanced visualization and interaction techniques such as magic lenses and bimanual interaction. Document sharing is also supported with single display groupware as well as remote shared editing. The paper describes the component model and illustrates the use of the toolkit through concrete examples, including multiple views and concurrent interaction.

This paper addresses and resolves an interesting question concerning the reason for anthropomorphic user interface feedback being more effective (in two of three contexts) and preferred by users compared to an equivalent non-anthropomorphic feedback. Firstly the paper will summarise the author’s three internationally published experiments and results. These will show statistically significant results indicating that in two of the three contexts anthropomorphic user interface feedback is more effective and preferred by users. Secondly some of the famous work by Reeves and Nass will be introduced. This basically shows that humans behave in a social manner towards computers through a user interface. Thirdly the reasons for the obtained results by the author are inextricably linked to the work of Reeves and Nass. It can be seen that the performance results and preferences are due to the subconscious social behaviour of humans towards computers through a user interface. The conclusions reported in this paper are of significance to user interface designers as they allow one to design interfaces which match more closely our human characteristics. These in turn would enhance the profits of a software house.

Agent technology has been suggested by experts to be a promising approach to fully extend Intelligent Tutoring Systems (ITS). By using intelligent agents in an ITS architecture it is possible to obtain an individual tutoring system adaptive to the needs and characteristics of every student. The general architecture of the ITS proposed is formed by the three components that characterize an ITS - the Student Model, the Domain Model, and the Education Model. In the Student Model the knowledge that the system has about the student (profile and interaction with the system) is represented. In the Domain Model the knowledge about the contents to be taught is stored. Precisely, in this model four autonomous agents - the Preferences Agent, the Accounting Agent, the Exercises Agent and the Tests Agent - have been defined. Lastly, the Education Model provides the functionality that the teacher needs. Across this module, the teacher changes his preferences, gives reinforcement to the students, obtains statistics and consults the matter.

Usability has become a critical quality factor of software systems in general, and especially important regarding Web-based applications. Measuring quality is the key to developing high-quality software, and it is widely recognised that quality assurance of software products must be assessed focusing on the early stages of the development process. This paper describes a controlled experiment carried out in order to corroborate whether the patterns associated to a quality model are closely related to the final Web application quality. The experiment is based on the definition of a quality model and the patterns associated to its quality criteria to prove that applications developed using these patterns improve its usability in comparison with other ones developed without using them. The results of this experiment demonstrate that the use of these patterns really improves the quality of the final Web application in a high degree. The experiment is formally based on the recommendations of the ISO 9126–4.