Seminars given for CSE700 Seminars Course;

Assc. Prof.  Ali Fuat Alkaya

Title: Migrating Birds Optimization: A New Meta-heuristic Approach and Its Application to the Quadratic Assignment Problem

Abstract:  We propose a new nature inspired metaheuristic approach based on the V flight formation of the migrating birds which is proven to be an effective formation in energy saving. Its performance is tested on quadratic assignment problem instances arising from a real life problem and very good results are obtained. The quality of the solutions we report are better than simulated annealing, tabu search, genetic algorithm, scatter search, particle swarm optimization, differential evolution and guided evolutionary simulated annealing approaches. The proposed method is also tested on a number of benchmark problems obtained from the QAPLIB and in most cases it was able to obtain the best known solutions. These results indicate that our new metaheuristic approach could be an important player in metaheuristic based optimization.

--------------

Asst. Prof.   Murat Can Ganiz

Title: Semantic Text Mining: Supervised and Semi-Supervised Classification Approaches*

Abstract:  Ever increasing volumes of structured and unstructured data in variety of types such text, multimedia, and graph is popularly referred as "Big Data".  An important portion of this data is in unstructured text format. There are important challenges to extract meaning or useful information from large textual datasets. One of the major methods used for organizing and mining big textual data is text classification. We use classification algorithms from machine learning domain which is a sub-field of Artificial Intelligence. In order for machine learning algorithms to work, unstructured textual documents are represented using Vector Space Models (VSM). Semantics of VSM comes from the distributional hypothesis, which states that words that occur in similar contexts usually have similar meanings. Although, the VSM  approach is very simple and commonly used, it has several limitations. One of them is the assumption of independency between terms as well as between documents. Documents are represented only with their term frequencies, disregarding their position in the document or their semantic or syntactic connections between other words. As a result, polysemous words (i.e. words with multiple meanings) are treated as a single entity and synonymous words are mapped into different components. This has a negative impact on performance of machine learning algorithms. There is a need for more advanced methods that incorporates additional semantics into these algorithms. In this seminar, we present several semantic text mining algorithms ranging from higher-order co-occurrence based probabilistic methods to class semantics based kernel methods for supervised and semi-supervised classification of textual documents. 

*The research work presented in this seminar has been supported in part by The Scientific and Technological Research Council of Turkey (TÜBİTAK) grant number 111E239.