DL-Learner 1.4.0
- Mailing list has moved from sourceforge to freelists.org
- DL-Learner system has been presented at The Web Conference in Lyon: DL-Learner – Structured Machine Learning on Semantic Web Data
- Dependency Updates
- OWLAPI: 4.5.13
- Jena: 3.12.0
- Hermit: 1.4.3
- Edge: 3.2
- CoreNLP: 3.9.2
- Compilation with Java 11, 12 fixed
- Tests now run on Windows
- Breaking changes;
- The Accuracy Methods were moved to
org.dllearner.accuracymethods
package
- The Accuracy Methods were moved to
- Added batch modes to Pattern and property learning algorithm
- Added minimumTreeScore in the disjunctive el learning algorithm
- Began work on new Class Expression Learning Problem
- Continued work on Query Tree Learning algorithm
- Improved CBD generators
- Improved Closed World Reasoner
- Improved Range based facet restrictions
- Improved simple SPARQL subclass learner
- Many utilities for working with JGraphT graphs
- New CELOE usage example
- New Examples Provider utility
- New JSON config file example
- SPARQL Reasoner adaptions for OntoQuad triple store
- Some logical fixes to the Horizontal Expansion in CELOE reported/analysed by Yingbing Hua
- Other bug fixes and improvements
DL-Learner 1.3
- Terminological Decision Tree and Evidence-based Terminological Decision Tree learning algorithms
- More information about TDT and ETDT in an upcoming EKAW publication: ‘Integrating New Refinement Operators in Terminological Decision Trees Learning’
- SPARQL Reasoner component has been evaluated in ECAI’16 publication: Towards SPARQL-Based Induction for Large-Scale RDF Data Sets
- Addition of the LEAP and EDGE systems and BUNDLE reasoner for probabilistic OWL
- Thanks to Giuseppe Cota from the University of Ferrara for the contribution
- New command line tool to run leap configurations: `clileap`
- Part of the extended DL-Learner release
- Query Tree Learning support for incoming edges
- Further QTL improvements
- New tree-based Concise Bounded Description generator
- Rho refinement operator now respects classes ignored in the learning algorithm
- For PosNegLP, positive and negative examples can be describes with a class expression
- Class Learning Problem support for pluggable accuracy methods
- New class expression length metric to further tune the refinement
- New expression validation client to calculate test coverage for PosNeg learning problems
- Windows drive letters are now supported in OWLFile loader
- An updated review of the DL-Learner framework has been published in the Journal of Web Semantics: DL-Learner—A framework for inductive learning on the Semantic Web
- Please refer to that publication when citing DL-Learner
- Framework updates
- Update to Java 8
- Update to Jena 3
- Update to OWL API 4.2
- Update to Spring 4.3
- API and deprecation notices
- getName() has been removed, use the AnnComponentManager instead
- ROLearner2 has been integrated into OCEL
- The refinement package has been removed
- name attribute of configOption has been removed
- Reasoner should implement getDomainImpl
- QueryEngineHTTP is now upstreamed
- CLIBase2 serves as a base for new CLI components
- The BeanXMLConverter has been removed
- Bug fixes
DL-Learner 1.2
- Direct (without reasoner) SPARQL based learning
- Extended Query Tree Learning (QTL) algorithm
- OWLAPI v4 support
- Spring v4 support
- Cleaned up deprecated methods
- Bug fixes
DL-Learner 1.1
- Preliminary support for XSD date/time literals
- Preliminary support for inverse properties
- Better support for SPARQL endpoints as knowledge sources
- New Query Tree Learning (QTL) algorithms
- Optional DL syntax output instead of Manchester
- Documentation updates
- Bug fixes
DL-Learner 1.0
- Project migrated to Github
- All internal data structures migrated to OWL API
- Protégé plugin moved to a separate project
- Several improvements in query tree learner for robustness against noisy input
- Experimental Punning Handling
- New Utilities
- Concise Bounded Description generator
- Jena-OWLAPI converter
DL-Learner 1.0 Beta-3 (2014-05-23)
- ISLE algorithm for combining ELTL with natural language input
- Pattern based enrichment algorithm
- ParCEL algorithm for a parallel divide & conquer strategy
- Better fragment extraction for large knowledge bases
- TBSL question answering algorithm moved to separate project
- Modifications for smoother integration of ELK and CEL reasoners
DL-Learner 1.0 Beta-2 (2012-04-06)
- Removed dependencies to Sesame and several updated dependencies including Jena
- Thread safety improvements and parallelism in several algorithms and processing steps
- Lightweight REST interface
DL-Learner 1.0 Beta-1 (2011-12-08)
- Added stochastic sampling for cases with many input examples to further algorithms
- Modification of conf file format and improved compatibility for many algorithms
- Significant work on using DL-Learner for feedback loops in question answering
- Many bug fixes
DL-Learner 1.0 Alpha-2 (2011-09-15)
- Bugfix release
- OWL API communication and command line interface issues resolved
DL-Learner 1.0 Alpha-1 (2011-09-03)
- Major refactoring of command line interface and conf files to internally build on Java Spring
- New conf file syntax introduced with much greater flexibility to create and configure objects (the configuration file syntax is internally mapped to Spring XML configuration objects, so it is essentially a more editing friendly variant of Spring XML)
- Wider support for different OWL axiom types in learning algorithms
- Preliminary support for fuzzy description logics
- Query tree learner
Build 2010-08-07
- Support for OWL API 3
- ORE tool based on DL-Learner algorithms (soon to be migrated to an own project)
- Implemented several new heuristics, e.g. generalised F-Measure
- Stochastic approximation of computing F-Measure
- Learning algorithms for the EL description logic
- Support for hasValue construct in combination with string datatype
- Support for refining existing definitions (instead of learning from scratch) for CELOE ontology engineering algorithm
- Increased number of unit tests (now 40)
- Support for direct Pellet 2 integration and reasoners connected via OWLLink
- 24 bugs fixed and 12 feature requests implemented at sourceforge.net bug tracker
Build 2009-05-06
- New algorithm: CELOE (class expression learning for ontology engineering)
- Protégé Plugin based on CELOE
- Wrote a manual for DL-Learner
- An efficient refinement operator for the EL description logic
- Fast stochastic class expression coverage estimation included
- Reasoner component design and learning problem structure improved
- More learning examples provided and unit tests for ensuring code quality extended
- 6 bugs and feature requests reported at the sourceforge.net tracker fixed
Build 2008-10-13
- Improved refinement operator based learning approach taking domain/range of properties, property hierarchies, disjoint classes into account to structure search space more efficiently
- DL-Learner GUI for loading, saving, and modifying configuration files
- Fast instance checking algorithm reduces the time to test example coverage of class descriptions significantly
- Carcinogenesis Benchmark
- Extraction component: more flexible structure, SPARQL results are converted to OWL on the fly, correct blank node handling Poster Abstract
- More learning examples provided in release
- 12 bugs and 10 feature requests reported at the sourceforge.net tracker fixed
Build 2008-02-18
- Flexible new component based structure:
- 4 types of components: knowledge sources, reasoners, learning problems, learning algorithms
- easily extensible: to implement a new component of one of the above types you only have to extend the corresponding class in org.dllearner.core and add the name of your class to the components.ini file
- each component can maintain and easily extend its own configuration options
- Support for using SPARQL endpoints as background knowledge, including mechanisms for knowledge fragment selection. This feature enables DL-Learner to use DBpedia as background knowledge.
- Preliminary support for learning from only positive examples and learning of inclusion axioms instead of definitions.
- Support for N-Triple files.
- Support for using role hierarchies in the refinement operator based algorithm.
- Much more powerful web service interface allowing to access and modify all DL-Learner components.
- Reasoners:
- preliminary OWL API reasoner interface support: Pellet, FaCT++
- KAON2 dropped, such that DL-Learner now depends solely on open source libraries
- A Prolog parser, which can help in converting Prolog files to OWL (thereby transfering ILP problems into OWL learning problems).
- More examples added:
- complete Moral Reasoner Benchmarks
- more SPARQL benchmarks
- all examples now also available in OWL
Build 2007-08-31
Initial release.