Radu Razvan Slavescu

A hybrid approach to QoS-aware Web Service classification and recommendation

The proliferation of Web Services (WSs) has given significant importance to the Quality of Service (QoS) in the discovery process. This paper introduces a hybrid system which combines semantic technology and logic based reasoning (i.e. OpenCyc) with numerical calculus to classify, evaluate and then recommend QoS aware WSs. The system is able to answer high level composed functional and/or nonfunctional queries related to a real-life dataset containing 2500 WS, therefore enabling classification and recommendation of WSs.

LELA - a natural language processing system for Romanian tourism

This paper presents a commercial semantic-based system for the Romanian tourism. The Lela system exploits both open linked data from Romanian and international sources, and also proprietary databases in the tourism domain. We present the process of creating the linked data set, based on: i) engineering the LELA Romanian tourism ontology, and ii) populating the ontology by linking open data. The system also provides a natural language interface for the Romanian language. The queries are automatically translated into SPARQL based on a controlled vocabulary derived from the Lela ontology.

Romanian2SPARQL: A Grammatical Framework approach for querying Linked Data in Romanian

SPARQL queries have become the current state of the art for querying linked open data repositories. Given the difficulties to create such queries by non-expert users, interfaces that accept questions in natural language and convert them to SPARQL are one solution to this problem. We developed a translator for Romanian natural language to SPARQL. The system exploits the conceptual power of the Grammatical Framework library and the Haskell programming language. The system is validated in the Romanian cultural domain against various Linked Data Sources, including DBpedia.

A model of trust and reputation in e-commerce based on multimodal logic

We use a multimodal logic to describe trust and reputation in a community of agents who interact, assess the quality of the result and possibly publish it. Axioms are added for dealing with reputation inside the group. The formalism implementation is tested in a multiple-round e-commerce scenario.

Association rules mining with categories and regular expressions based pruning

We present an extended version of the well known Apriori algorithm used for mining association rules. Its novelty consists in employing regular expressions in order to specify the form of the rules intended to be generated. These are further used for pruning in the rule generation step, thus speeding up the process. We show that the new version of the algorithm was used for generating medical hypotheses for some problems in the field without risking to loose generality.

Building annotation rules for text description of endoscopies in romanian - an NLP-free approach

This paper presents an approach for building annotation rules for texts containing natural language descriptions of the endoscopies, in Romanian. The annotation rule extraction relies on running the Apriori algorithm over some previously annotated texts. It does not employ any Natural Language Processing tools. We show by experiments that, for the relatively small vocabulary employed by those descriptions, the solution proves to offer quite good performance.

Designing an Information Gathering Application for a Personalized Travel Recommender System

Obtaining information which is both trustworthy and relevant to the user is crucial for the success of the Travel Recommender Systems. Such a system should automatically obtain high quality facts and knowledge about the touristic attractions, filter them and present the user only those pieces which are of interest for him. To this end, a challenging issue is to select among the existing technologies those able to be articulated into a system which could fulfill the mentioned task. The system we propose has a pragmatic approach of the problem, aiming to identify the instruments to be used in touristic information gathering, if integrated in such a system. The system is built around a map system and a general purpose ontology and makes use of parsing techniques in order to retrieve additional information from web sources via a search engine. The proposed architecture is also expandable and other technologies can be linked to improve the variety of results.

Helpful Behavior Based on Trust for Web Services

As non-functional properties of composite software services involve a good deal of uncertainty, we consider them as shared goals for an agent team. An architect is focused on choosing each service and its provider, while the consultant will suggest, when help is possible, an alternative so that the likelihood of success is maximized. We employ a structure called”Belief Recipe Tree” for supporting this kind of decisions and build the necessary beliefs starting from a set of prior experiences the consultant has had with different provider and/or abstract plans. This approach is shown to provide a higher flexibility due to the fact that beliefs are revised after each experience.

On a diacritics due part-of-speech tagging ambiguity in Romanian

In this paper we investigate lack-of-diacritics due ambiguity when doing Part-of-Speech tagging in Romanian. This means that some words, if they are written with no diacritcs, could have associated more than one Part-of-Speech tag, even if this is not the case when the diacritcs are employed. A method for dealing with this problem is proposed. By developing an existing Hidden Markov Model and employing a large dictionary and trigrams set, the solution allows doing Part-of-Speech tagging for Romanian even if the words are written with no diacritics. The performance is evaluated on a set of sentences manually built to illustrate the situation.

Enhancing digital maps to support reasoning on traffic sign compliance

This article presents a new methodology of enriching digital maps in order to reason upon the traffic sign compliance, of traffic participants. The Open Street Map (OSM) data is augmented with a new layer for road orientation and road intersections in order to support not only map matching, but also reasoning processes on the compliance to traffic regulations provided by traffic signs. In order to increase the accuracy of the road map, traces collected with a high precision GPS receiver are used in a probabilistic model of traffic lanes. A distribution of points across the road relative to the existing OSM center-line is obtained by learning Gaussian Mixture Models. A method for extracting accurate position for static points is also proposed and applied for positions of traffic signs. The enhanced digital map is applied on the identification of cars percepts which support the rule based reasoning about traffic sign compliance.