Vol. 28 · No. 4

Personalization of web-based information systems based on specialized user models has become more important in order to preserve the effectiveness of their use as the amount of available content increases. We describe a user modeling approach based on automated acquisition of user behaviour and its successive rule-based evaluation and transformation into an ontological user model. We stress reusability and flexibility by introducing a novel approach to logging, which preserves the semantics of logged events. The successive analysis is driven by specialized rules, which map usage patterns to knowledge about users, stored in an ontology-based user model. We evaluate our approach via a case study using an enhanced faceted browser, which provides personalized navigation support and recommendation.

We present a novel method for instance comparison of ontological concepts with regard to personalized content presentation and/or navigation in large information spaces. We assume that comparing properties of documents which users found interesting leads to discovery of information about users' interests specifically when considering Semantic Web applications where documents or their parts are represented by ontological concepts. We employ the ontology structure and different similarity metrics for datatype and object properties and investigate reasons behind user interest in the presented content. Moreover, we propose and evaluate an approach to instance similarity computation for a particular user while also considering the user's individual preferences.

This paper focuses to a formal model of user preference learning for content-based recommender systems. First, some fundamental and special requirements to user preference learning are identified and proposed. Three learning tasks are introduced as the exact, the order preserving and the iterative user preference learning tasks. The first two tasks concern the situation where we have the user's rating available for a large part of objects. The third task does not require any prior knowledge about the user's ratings (i.e. the user's rating history). Local and global preferences are distinguished in the presented model. Methods for learning these preferences are discussed. Finally, experiments and future work will be described.

Searching top-k objects for many users face the problem of different user preferences. The family of Threshold algorithms computes top-k objects using sorted access to ordered lists. Each list is ordered w.r.t. user preference to one of objects' attributes. In this paper the index based methods to simulate the sorted access for different user preferences in parallel are presented. The simulation for different domain types – ordinal, nominal, metric and hierarchical – is presented.

We present models, methods, implementations and experiments with a system enabling personalized web search for many users with different preferences. The system consists of a web information extraction part, a text search engine, a middleware supporting top-k answers and a user interface for querying and evaluation of search results. We integrate several tools (implementing our models and methods) into one framework connecting user with the web. The model represents user preferences with fuzzy sets and fuzzy logic, here understood as a scoring describing user satisfaction. This model can be acquired with explicit or implicit methods. Model-theoretic semantics is based on fuzzy description logic mboxf- mathcalEL. User preference learning is based on our model of fuzzy inductive logic programming. Our system works both for English and Slovak resources. The primary application domain are job offers and job search, however we show extension to mutual investment funds search and a possibility of extension into other application domains. Our top-k search is optimized with own heuristics and repository with special indexes. Our model was experimentally implemented, the integration was tested and is web accessible. We focus on experiments with several users and measure their satisfaction according to correlation coefficients.

Automated annotation of web documents is a key challenge of the Semantic Web effort. Semantic metadata can be created manually or using automated annotation or tagging tools. Automated semantic annotation tools with best results are built on various machine learning algorithms which require training sets. Other approach is to use pattern based semantic annotation solutions built on natural language processing, information retrieval or information extraction methods. The paper presents Ontea platform for automated semantic annotation or semantic tagging. Implementation based on regular expression patterns is presented with evaluation of results. Extensible architecture for integrating pattern based approaches is presented. Most of existing semi-automatic annotation solutions can not prove it real usage on large scale data such as web or email communication, but semantic web can be exploited only when computer understandable metadata will reach critical mass. Thus we also present approach to large scale pattern based annotation.