Integrating Feature Analysis and Background Knowledge to Recommend Similarity Functions

Existing approaches in similarity analysis is little concerned with the right choice of similarity functions. We present an approach for suggesting which similarity functions (e.g., edit distance) are most appropriate for a given similarity search task. We identify data features (e.g., misspellings) that are considerable when choosing similarity functions. We also introduce the concept of similarity function background knowledge that associates data features with similarity functions, and apply the knowledge to recommend suitable similarity functions.