2014 | OriginalPaper | Buchkapitel
Transition-Sensitive Distances
verfasst von : Kaoru Yoshida
Erschienen in: Similarity Search and Applications
Verlag: Springer International Publishing
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
In information retrieval and classification, the relevance of the obtained result and the efficiency of the computational process are strongly influenced by the distance measure used for data comparison. Conventional distance measures, including Hamming distance (HD) and Levenshtein distance (LD), count merely the number of mismatches (or modifications). Given a query, samples mapped at the same distance have the same number of mismatches, but the distribution of the mismatches might be different, either disperse or blocked, so that other measures must be cascaded for further differentiation of the samples. Here we present a new type of distances, called transition-sensitive distances, which count, in addition to the number of mismatches, the cost of transitions between positionally adjacent match-mismatch pairs, as part of the distance. The cost of transitions that reflects the dispersion of mismatches can be integrated into conventional distance measures. We introduce transition-sensitive variants of LD and HD, referred to as TLD and THD. It is shown that while TLD and THD hold properties of the metric similarly as LD and HD, they function as more strict distance measures in similarity search applications than LD and HD, respectively.