Link Search Menu Expand Document

Matcher Evaluation Core Concepts

Evaluator

An evaluator is an abstract class for creating any evaluation based on an ExecutionResultSet.

ExecutionResultSet

An ExecutionResultSet is the result of an Executor. It represents a collection of individual ExecutionResults.

Executor

An Executor runs a matcher or a list of matchers on a single test case or a list of test cases. You can run any matcher class (your own implemenation) but also packaged matchers by wrapping them as matchers (for details see the sections on Matcher Packaging).

Filter

A filter is a matcher that does not add new correspondences to the alignment but instead further processes the given alignment by (1) removing correspondences and/or (2) adding new feature weights to existing correspondences. MELT filters implement the Filter interface. Filters are intended to be used in a matcher pipeline.

GoldStandardCompletness

Most test cases are complete, i.e. correspondences that are not in the gold standard are regarded as wrong. However, if you have a (local) test case that is not complete, you can provide the level of gold standard completeness via enum GoldStandardCompleteness. The default evaluators will later on correctly evaluate system alignments for incomplete or partially complete test cases. Please refer to the API documentation to learn about the individual levels of completeness that are supported in MELT.

Matcher

A matcher is a process which, given two ontologies and an alignment, returns an alignment. Matchers can be used on their own or can be combined in a modular way through a matching pipeline. MELT offers various abstract matcher classes that can be extended such as MatcherYAAA, MatcherYAAAJena, or MatcherYAAAOwlApi.

TestCase

A TestCase is an individual matching task that may belong to a Track. A test case consists of two URIs for source and target ontology/graph as well as an URI to the reference (also known as solution or gold standard of the matching task).

Track and TrackRepository

A track is a collection of multiple TestCases. A track may be available online via the MELT TrackRepository.