Releases: dice-group/gerbil
Releases · dice-group/gerbil
GERBIL 1.2.3
Features
- Decreased the time needed to load the most recent results (#77)
Bug Fixes
- Fixed the problem that uploaded NIF-file-based datasets did not have been initialized using the EntityChecker and SameAsRetriever mechanism.
- Fixed the AIDA/CoNLL Dataset adapter. (#118)
- Fixed a bug that prevented the correct loading of a dataset if more than one annotator should be tested.
- Fixed the RDF representation of experiment results.
Known Problems
- Because of the retrieving of sameAs relations as well as the checking of all entity URIs, the evaluation takes more time than in older versions.
GERBIL 1.2.2
Features
- Implemented a better dataset handling (#66).
- Implemented the checking of entities (#98).
- Implemented a cache for the entity checking that stores checks for 1 month to increase the performance.
- Implemented the GSInKB measures
- Added a domain based sameAs retrieval manager that can call different sameAs retrievers based on the domain of a URI.
- Added a sameAs retriever that serves as a bridge between Wikipedia and DBpedia.
- Added certain domains to the properties file for which the HTTP-based same as retriever is used preventing the system from performing unnecessary HTTP requests.
- Added a SameAsRetriever handling the URI encoding problem.
- Added a sameAs retriever decorator that implements a crawling strategy instead of asking only once for the first given URI.
- Refactored the evaluation by implementing a classification of the markings that is done before the main evaluation starts. The confidence scores are considered more directly inside the evaluation at the stage of the F-measure calculation. Thus, the matching counting is only repeated for the single documents with the different confidence values present inside these documents instead of repeating it several times for the complete corpus.
- Added the FREME NER annotator configuration (#109).
- Added the FRED annotator configuration (#112).
- Added a client for the AIDA web service (#107).
Bug Fixes
- Fixed a bug inside the DBpedia Spotlight client.
- Fixed a bug that disabled the drawing of the SVG diagrams in Chrome-based browsers.
- Fixed a bug in the configuration page (#100).
- Updated the start.sh script.
- Fixed the problem of unmerged MSNBC entities (#103).
- Fixed a bug in the time measuring that was caused by the fact, that a single annotator instance could have been blocked by a Semaphore while the time measurement already started (#108).
- Fixed tooltips in the front end that haven't been shown correctly.
- Added the gerbil.nif.transfer library version 1.2.1 that fixes communication problems with the FREME NER annotator.
Known Problems
- Because of the retrieving of sameAs relations as well as the checking of all entity URIs, the evaluation takes more time than in older versions.
Version 1.2.1 of the gerbil.nif.transfer library
This is the fourth release of the gerbil.nif.transfer library. This simple lib translates a document from java into NIF for being transfered to a webservice or client and retranslates it into a java object. It is part of the GERBIL project and can be used by others to implement a NIF based web service for their own annotators.
- Added clone methods
- Implemented a common behaviour regarding the setting of
Set<String>attributes - All
Spanimplementations that are translated into NIF nodes additionally get the typenif:Phrase
GERBIL 1.2.1
Changes from 1.2.0 to 1.2.1
- Added a filter to the D2KB and Entity Typing experiment tasks that are removing all markings that do not (strongly) match the positions of the mentions that have been given in the request (refs #98).
Changes from 1.1.4 to 1.2.0
Features
- Contains a new benchmark core that is based on URIs instead of Wikipedia IDs (refs #51 and #74). Thus, the evaluation is knowledge base agnostic (refs #88).
- Implements a better URI matching. Added the concept of well-known knowledge bases to ensure the identification of emerging entities (solves #94).
- The evaluation is more flexible and open to new metrics (refs #5).
- Added Task 1 as well as Task 2 of the OKE Challenge as new experiment types.
- Added Entity Typing (#48) as experiment type and a hierarchical f-measure for its evaluation.
- Added the new concept of sub tasks. Thus, it can be seen whether the linking or the recognition of an annotator caused the most problems.
- The single responses of the annotators can be stored in a NIF file for later debugging.
- Offers backwards compatibility to the BAT-framework
- Gives a better way to activate and deactivate single annotators and datasets that is based on property files instead of coded java classes (refs #22).
- Added run time measurement (refs #63).
Bug Fixes
Version 1.2.0 of the gerbil.nif.transfer library
This is the third release of the gerbil.nif.transfer library. This simple lib translates a document from java into NIF for being transfered to a webservice or client and retranslates it into a java object. It is part of the GERBIL project and can be used by others to implement a NIF based web service for their own annotators.
- Added scored spans
GERBIL OKE2015
This is an adapted version of GERBIL that has been used for the OKE Challenge 2015.
Features:
- Contains a new benchmark core that is not based on the BAT-Framework but on URIs instead of Wikipedia IDs (refs #51). Thus, benchmarking the linkage to other knowledge bases than Wikipedia as well as the identification of emerging entities is now supported.
- The evaluation is more flexible and open to new metrics (refs #5).
- Added the concept of well-known knowledge bases to ensure the identification of emerging entities.
- Added Task 1 as well as Task 2 of the OKE Challenge as new experiment types.
- Added Entity Typing as experiment type and a hierarchical f-measure for its evaluation.
- Added the new concept of sub tasks. Thus, it can be seen whether the linking or the recognition of an annotator caused the most problems.
- Used version 1.1.0 of the gerbil.nif.transfer library and its more detailed datatype hierarchy.
- Compared to the last version of GERBIL, this adapted version has not all features of the "normal" GERBIL since they are not needed for the OKE Challenge 2015 and their adaption to the new benchmarking core is still work in progress.
Missing Features:
- C2KB, Rc2KB, Sc2KB and Sa2KB are missing in this release
- Some adapters (annotators as well as datasets) might be missing
Version 1.1.0 of the gerbil.nif.transfer library
This is the second release of the gerbil.nif.transfer library. This simple lib translates a document from java into NIF, transfers it to a webservice or client and retranslates it into a java object. It is part of the GERBIL project and can be used by others to implement a NIF based web service for their own annotators.
- Created a structure of classes making the library more modular and, thus, more reusable
- Created a clearer hierarchy of markings (see the wiki for details)
- Extended the Meaning interface to be able to have more than one single URI
- Added a type inference supporting different ways to express the type of a named entity
- Extended the IO part of the library
GERBIL 1.1.3
Features:
- Added the calculation of correlations of dataset features and annotator performance.
- Added the new version of Babelfy (pre-release version 1.0). This version does not need a local instance of BabelNet.
- Added the usage of a small worker library that eases the observing of single tasks.
Bug Fixes:
- Prepared a better handling of the time measurement. But it is not used by the system, yet. (refs issue #63)
- Added a new URL for the DEXTER web service
- Removed the usage of the BAT-Framework's internal cache that caused large problems in combination with GERBIL's structure.