Catálogo de publicaciones - libros

Most research on social networks is concerned with information transmission per se Our aim here is to supplement the social network perspective by incorporating mechanisms that govern social influence Research in social psychology suggests that individuals interact, in large part, to construct a shared reality that consists not only of shared information but also of agreed upon opinions. In this process, they do not simply transmit information, but more importantly, they influence one another to arrive at a common interpretation of information. We will discuss similarities and differences in how networks structure shapes the spread of information and governs social influence. Both simulation and empirical data concerning these two processes show that they operate in a very different way. The spread of information, described as a contagion process describes how individuals learn about new facts. Social influence process describes how individuals evaluate and weight different items of information and how they change their opinions and attitudes. The results of numerous experiments have shown that three critical factors determine the impact of social influence: (1) the number of sources exerting the influence, (2) the immediacy of the source(s) to the target(s), and (3) the strength of the source(s). The process by which humans construct social reality may prove informative for designing rules of interaction among intelligent agents. The primary implication of the present model is that information is not merely acquired, but also evaluated and negotiated in a social context. The process by which humans evaluate information and construct social reality may prove informative for designing rules of interaction among intelligent agents. The primary implication of the present model is that information is not merely acquired, but also evaluated and negotiated in a social context.

AI systems must be able to learn, reason logically, and handle uncertainty. While much research has focused on each of these goals individually, only recently have we begun to attempt to achieve all three at once. In this talk I will describe Markov logic, a representation that combines first-order logic and probabilistic graphical models, and algorithms for learning and inference in it. A knowledge base in Markov logic is a set of weighted first-order formulas, viewed as templates for features of Markov networks. The weights and probabilistic semantics make it easy to combine knowledge from a multitude of noisy, inconsistent sources, reason across imperfectly matched ontologies, etc. Inference in Markov logic is performed by weighted satisfiability testing, Markov chain Monte Carlo, and (where appropriate) specialized engines. Formulas can be refined using inductive logic programming techniques, and weights can be learned either generatively (using pseudo-likelihood) or discriminatively (using a voted perceptron). Markov logic has been successfully applied to problems in entity resolution, social network modeling, information extraction and others, and is the basis of the open-source Alchemy system.

(Joint work with Stanley Kok, Hoifung Poon, Matt Richardson and Parag Singla.)

We describe a new knowledge acquisition tool that enabled us to develop a dialog system recommending software design patterns by asking critical questions. This assistance system is based on interviews with experts. For the interviews we adopted the repertory grid method and integrated formal concept analysis. The repertory grid method stimulates the generation of common and differentiating attributes for a given set of objects. Using formal concept analysis we can control the repertory grid procedure, minimize the required expert judgements and build an abstraction based hierarchy of design patterns, even from the judgements of different experts. Based on the acquired knowledge we semi-automatically generate a Bayesian Belief Network (BBN), that is used to conduct dialogs with users to suggest a suitable design pattern for their individual problem situation. Integrating these different methods into our knowledge acquisition tool KARaCAs enables us to support the entire knowledge acquisition and engineering process. We used KARaCAs with three design pattern experts and derived approximately 130 attributes for 23 design patterns. Using formal concept analysis we merged the three lattices and condensed them to approximately 80 common attributes.

As new semantic web standards evolve to allow quantified rules in FOL, we need new ways to capture them from end users in RDFS(XML). We show how to do this against a graphic view of Entities and their Relationships (associated or derived). This even allows inclusion of existential quantifiers in readable fashion. The captured constraint can be tested by generating queries to search for violations in stored data. The constraint can then be automatically revised to exclude specific cases picked out by the user, who is spared worries about proper syntax and boolean connectives.

Constraint satisfaction is a powerful approach to solving a wide class of problems. However, as many non-experts have difficulties formulating tasks as Constraint Satisfaction Problems (CSPs), we have built a number of interfaces for particular kinds of CSPs, including crypt-arithmetic problems, map-colouring problems, and scheduling tasks, which ask highly focused questions of the user, c.f., the earlier MOLE/MORE, and SALT knowledge acquisition systems. Information from each of these interfaces is then transformed initially into a structured format which is semantic web compliant and is secondly transformed into the format required by the generic constraint satisfaction problem solver. When this problem solver is run, the user is either provided with solution(s) or feedback that the problem is underspecified (when many solutions are feasible) or over-specified (when no solution is possible). The system has 3 distinct phases, namely; information capture, transformation of the information to that used by a standard problem solver, and thirdly the solving and user feedback phase.

Evaluation of knowledge acquisition (KA) is difficult in general because of the costs of using a human expert. In this paper, we use a general simulation framework to evaluate some aspects of KA. We focus on the importance of acquiring domain ontological structures and the use of stored or cornerstone cases to validate changes. We find that the for a higher level of expertise, an ontology is very useful, but cornerstone cases less so, but the weaker the level of expertise, the more valuable the cornerstone cases and the less helpful an ontology.

A predictive mathematical model is presented for the expected case-driven transfer of classification rules. Key insights are offered for Knowledge Acquisition in expert systems, machine learning, artificial intelligence, ontology, and folksomonies.

The main problem encountered when starting verification of goals for some formal system, is the ambiguity of those goals when they are specified in natural language. To verify goals given in natural language, a translation of those goals to the formalism of the verification tool is required. The main concern is to assure equivalence of the final translation and the original. A structured method is required to assure equivalence in every case.

This article proposes a goal formalisation method in five steps, in which the domain expert is involved in such a way that the correctness of the result can be assured. The contribution of this article is a conceptual goal model, a formal expression language for this model, and a structured method which transforms any input goal to a fully formalised goal in the required target formalism. The proposed formalisation method guarantees essential properties like , , and .

Reuse has long been a major goal of the Knowledge Engineering community. The focus of this paper is the reuse of domain knowledge acquired for an initial problem solver, with a further problem solver. For our analysis we chose a knowledge base system written in CLIPS based on the propose-and-revise (PnR) problem solver, and which had a lift/elevator knowledge base (KB). Given the nature of the problem solver, the KB contained 4 components, namely an ontology, procedural statements which specify how the artifact, the lift, could be enhanced/modified, a set of constraints to be satisfied, and a set of fixes to be applied when constraint violations occurred. These 4 components were first extracted manually, and were used with both an Excel spreadsheet and a constraint problem solver (ECLiPSe) to solve a range of tasks. The next phase was to implement ExtrAKTor which extracts the same 4 knowledge sources virtually automatically from the CLIPS knowledge base (held by Protégé), and transforms these so that they are usable with a number of problem solvers. To date Excel & ECLiPSe have been selected, and again we have demonstrated that the resulting systems are able to solve a variety of lift configuration tasks. This is in contrast to earlier work which produced abstract formulations of the problem but which were unable to perform reuse of actual knowledge bases.

The establishment of a computational framework of roles contributes effectively to the management of instance models because it provides us with a useful policy for treatment of views and contexts related to roles. In our research, we have developed an ontology building environment, which provides a framework for representation of roles and their characteristics. In this paper, as an extension of the framework, we present a framework for organizing roles according to their context dependencies. We especially focus on defining and organizing compound roles, which depend on several contexts.