Combining Index Structures for application-specific String Similarity Predicates

This paper presents new approaches for supporting string similarity matching based on a combination of techniques from the fields of information technology and computational linguistics to achieve better results regarding accuracy and efficiency. The homogenization of plain text reduces the volume of index structures and concurrently increases the quality of hit-lists. Furthermore it shows the careful and context dependent dealing with abbreviation, acronyms, and synonyms. The core of this work is a general approach to support stepwise application-specific and index supported string similarity predicates. It uses the Hybrid-Ternary Search Trie as one of the fastest index structure for strings. Tries guarantee best results for exact matching as well as inexact matching in preprocessed data and can be used for external data storage. Especially Hybrid-Ternary Search Tries are easy adaptable for common strings and provide best average results for inexact matching without any limitations.