Using transformations to improve semantic matching

Abstract

Many AI tasks require determining whether two knowledge representations encode the same knowledge. Solving this matching problem is hard because representations may encode the same content but differ substantially in form. Previous approaches to this problem have used either syntactic measures, such as graph edit distance, or semantic knowledge to determine the "distance" between two representations. Although semantic approaches outperform syntactic ones, previous research has focused primarily on the use of taxonomic knowledge. We show that this is not enough because mismatches between representations go largely unaddressed. In this paper, we describe how transformations can augment existing semantic approaches to further improve matching. We also describe the application of our approach to the task of critiquing military Courses of Action and compare its performance to other leading algorithms.

Theories of meaning in schema matching: An exploratory study

“…Evaluation of schema matching methods is done by comparing their results to a set of reference matches [3][4][5][6][7][8][9][10][11][12]. However, none of the experimental studies explicate the theories of meaning held by subjects or researchers that establish reference matches.…”

“…Experimental evaluations frequently suffer from lack of validity. When reference matches are established by a single domain expert, as in [4,32], the method and its performance may reflect the idiosyncrasies of the specific expert and may not be generalizable. When the researchers themselves [6,7,29,34,35] establish the reference matches, it is not surprising that the method achieves a high level of performance, as the method developers have intricate knowledge of the theory of meaning used to establish the reference matches.…”

“…Linguistic information may be used by measuring character string similarity [3,5,14,15,18] or by using externally supplied dictionaries, thesauri or lexical databases [4,11,16,19,[21][22][23][24][25] such as WordNet [26] or CyC [27]. Such linguistic databases provide lists of homonyms, synonyms, and other standard semantic relationships between words.…”

“…the relationships between database elements such as relationship-types between entity-types or foreign-key dependencies between tables [4,5,14,15,18,19,23,24,28]. The use of structural information for identifying matching database elements may be limited to local structures, where relationships only between directly connected database elements are considered, or it may encompass global structures, where the overall structure of the database is considered [14].…”

“…In these cases, all the information available is that in the database itself, primarily its schema and instances. The fact that all method performance evaluations show often significantly less than 100% success at matching [3][4][5][6][7][8][9][10][11][12], combined with the ongoing research effort in the field, shows that the problem is not yet completely solved.…”