melt-3.2
New
- introduced a new matcher:
SentenceTransformersMatcher
- example for track validation (used to analyze biodiv track)
- example for multi source matching called
multisource experiment
- eval client is build with every commit and the final jar is uploaded as a build artifact
- a nightly link is generated which always points to the latest version
- eval client now also works with short "built-in" track string names (e.g.
--track conference
) - all "built-in" tracks can be displayed in the evaluation client using option
--show-tracks
Fixed
- normalization of literals in
BaselineStringMatcher
(fixes issue #44) - matcher name was not always encoded (because it was used in as file name) (see issue #128)
- use a new local track for a local test case (fixes issue #121)
Improvements
- Web frontend for Web/Docker packaged matchers works now with javascript and provides better UX
- Server for Web/Docker based images now more stable
MatcherDockerFile
automatically logs the text which is logged in container - this helps to debug errors- the function
logAllLinesFromContainer
is thus deprecated
- the function
MatcherDockerFile
extracts the name for the image from the file content. Thus, the file which contains the image
can be named arbitrarily. The file compression is also automatically detected. See issue #123- documentation about jena interfaces like
LiteralExtractor
andTextExtractor
- welcome message also shows the git commit hash which is used to build the eval client
New tracks
- new biodiv track (version
2021
)