Similarity Judgement of Sequences by Incorporation of Structural Information

<<<back

Dissertation Project at the University of Osnabrück by Klaus Dalinghaus

ABSTRACT

The problem of judging the similarity of sequences is often solved with the dynamic programming approach. This method is based on a similarity measure or cost function judging the similarity of sequence elements and thus assumes the context independence of the elements. In some application domains, especially from the human perception like music or language, this context independence is not given. However, there is a knowledge about the structure in these sequences. Based on the structural alignment framework by Markman and Gentner, I present an algorithm for sequence alignment incorporating structural information. As an example domain I chose the recognition of rhythm patterns. The knowledge about the modelling of a domain often is incomplete or uses vague concepts. Hence, my approach is based on the modelling of the domain in fuzzy logic and optimising the resulting weights by learning from concrete examples. In this respect, my approach accounts for the improvement of the domain knowledge.

-- KlausDalinghaus -- 13 Apr 2006