Knowledge Engineering Research Group (KEG)

Led by: Rodica POTOLEA
Fields: Math. Comp. Sci. & Software Eng.
Institution: Technical University of Cluj-Napoca (UTCN)
Scientists/Total: 3/6
Keywords:  Context-sensitive content retrieval and (dis)similarity measures-extracting content, Supervised/semi-supervised/unsupervised machine learning techniques, Text mining, Social networks mining, Opinion mining and Sentiment analysis, Data integration.
Presentation

Context-sensitive content retrieval and (dis)similarity measures-extracting relevant content for a given context; data sources: free-text documents or data gathered via measuring/sensors/investigations. Main methods: lexical and/or syntactical analysis, ontology-driven concept identification, query expansion, dealing with negation and/or (in)consistencies, (dis)similarity measures.
Supervised/semi-supervised/unsupervised machine learning techniques–dealing with fundamental learning techniques, their current trends and applications, and the specific instantiations for very large, noisy, incomplete, imbalanced, and/or unstructured data.
Text mining – extracting knowledge from unstructured text via text processing methods, taxonomy and/or ontology support, learning techniques and domain specific information.
Social networks mining – deriving social relationships and behaviours from existing explicit or implicit connections. Identifying clusters, membership relations, habits, social patterns, trends and needs for individuals belonging to a group.
Opinion mining and Sentiment analysis– extracting opinions from reviews and/or other unstructured free-text; dealing with deceptive content, negation. Identify trendsin different groups from their indirect opinions.
Data integration – integrating heterogeneous data by ontology-driven, (semi-) automatic design of unified data structures and automatic design of the corresponding ETL processes.

Infrastructures

------

Applications

Context-sensitive content retrieval and (dis)similarity measures
Supervised/semi-supervised/unsupervised machine learning techniques
Text mining
Social networks mining
Opinion mining and Sentiment analysis
Data integration
Community detection
Schema mapping and data fusion
Decision support systems

Nokia prototype
We designed and developed a three-layered matching model that aims to identify the most relevant content in a given
context. Our model consists of three components: a lexical, a semantic, and a user-profile component and the
corresponding composed similarity metrics. We instantiated our model for the online advertisement problem and the
preliminary evaluations showed promising results in terms of information relevance.

SEARCH prototype
User profiling under dynamic environment: identify the profile of each individual interacting with the system, based on
initial (static) features and the interaction with the system (dynamic features) and providing learning content according to
the actual profile. The user classification module – which has been implemented by means of a SOM –allows for a
seamless transition from the initial static to the static and dynamic user assessment.

GridMOSI prototype
Benchmarking numerical algorithms: - parallel/distributed deployment of a set of numerical algorithms frequently
employed in applications.

Contact

Str. G.Baritiu Nr. 26-28, 400027
Cluj-Napoca, jud. Cluj, România
Tel: +40 264 202389 +40 264 402387
Fax: +40 264 594491
eMail: rodi...@cs.utcluj.ro
miha...@cs.utcluj.ro

 

 

 

0 Patents or Software

 

Research For Industry - Venture Contect Research For Industry - Arott Research For Industry - Romanian Startups Research For Industry - CCifer Research For Industry - Spherik Research For Industry - CIDO Research For Industry - Spherik