Computational collective intelligence

Computational Collective Intelligence (CCI) is most often understood as an AI sub-field dealing with soft computing methods which enable making group decisions or processing knowledge among autonomous units acting in distributed environments. Web-based systems, social networks and multi-agent systems very often need these tools for working out consistent knowledge states, resolving conflicts and making decisions.

Knowledge management systems¹

Knowledge Management System (KM System) refers to a (generally IT based) system for managing knowledge  in organizations for supporting creation, capture, storage and dissemination of information. It can comprise a part (neither necessary or sufficient) of a Knowledge Management initiative. The idea of a KM system is to enable employees to have ready access to the organization's documented base of facts, sources of information, and solutions. For example a typical claim justifying the creation of a KM system might run something like this: an engineer could know the metallurgical composition of an alloy that reduces sound in gear systems. Sharing this information organization wide can lead to more effective engine design and it could also lead to ideas for new or improved equipment.

¹ from Wikipedia

Agents and multi-agent systems

Agents and multi-agent systems are related to the modern software which has long been recognized as a promising technology for constructing autonomous, complex and intelligent systems. A key development in the field of agent and multi-agent systems has been the specification of agent communication languages and formalization of ontologies. Agent communication languages are intended to provide standard declarative mechanisms for agents to communicate knowledge and make requests of each other, whereas ontologies are intended for conceptualization of the knowledge domain. In this paradigm cognitive agents of heterogeneous nature possess diverse conceptual views and ontologies the problem of semantic mismatch arises, and a special conflict resolution strategies based on computer-supported negotiation are necessary.

Recommendation and personalization in web systems

The main aim of the recommendation systems is to deliver customized (personalized) information to a very differentiated users. They may be applied in a great variety of domains, such as: net-news filtering, web recommender, personalized newspaper, sharing news, movie recommender, document recommender, information recommender, e-commerce, purchase, travel and store recommender, e-mail filtering, music recommender, student courses recommender, user interface recommendation, negotiation systems, etc.. We consider two dimensions of the recommendation systems, user modeling and user model exploitation. The former considers user profile representation&maintenance and profile learning techniques. The later contains information filtering method, matching techniques and profile adaptation technique.

Ensemble and hybrid models of computational intelligence

Ensemble learning is a type of machine learning that studies algorithms and architectures that build collections, or ensembles, of statistical classifiers/regressors that are more accurate than a single classifier/regressor. This technique combine the output of machine learning algorithms, called “weak learners”, in order to get smaller prediction errors (in regression) or lower error rates (in classification). The individual estimator must provide different patterns of generalization, thus in the training process diversity is employed. Otherwise, the ensemble would be composed of the same predictors and would provide as good accuracy as the single one. It has been proved that the ensemble performs better when each individual machine learning system is accurate and makes errors on different examples. To the methods of ensemble learning we may include bagging, boosting, stacking, subsampling, random subspaces, mixture of experts, and others. 

Semantic Information Retrieval

Traditional Information Retrieval (IR) methods are roughly adequate for modern Web search and analysis. We focus on IR methodologies for current (Web 2.0) or even future (Web 3.0) search and analysis engines. The techniques range from link structure analysis to using social network relationship semantics. We use and research paradigms and technologies like:

• Semantic Web (OWL)
• Linked Data (RDF. SPARQL)
• Web ontologies (FOAF)
• Web data aggregation

Multimedia Information Processing

Since its early days hypertext has been used in association with multimedia (hypermedia), therefore different types of multimedia information are key ingredients of Web-based information systems. Our research covers the following aspects of the information processing:

• Audio signal processing
• Image recognition and video clustering
• Lossy and lossless compression
• Platform independent playback support

System Performance Analysis and Improvement

System performance and responsiveness are usually crucial issues for users, especially in Web environment. Constant system development should always be led in parallel with performance analysis. Our research in the field covers:

• Content caching techniques
• Usability testing
• Content indexing algorithms
• Web-based optimization techniques and best practices

E-Learning Methodologies

Modern e-learning (2.0) focuses mostly on Computer Supported Collaborative Learning (CSCL). Using Moodle (StOPKa3) as a primary tool for teaching is a great incentive for us for exploring new techniques and applications of online collaboration. Research areas in this field include:

• Applications of online collaboration paradigms, like wiki and data (video) conferencing
• Learning Management Systems (LMS) and Learning Content Management Systems (LCMS)
• Digital documentation techniques, like screencast (with a “talking head” window) and annotated (narrated) screenshot slides
• Examining based on real-time quizzes