0.9.0

What's Changed

Added

• Conflict of interest and author contributions statement extraction in header and segmentation models #1319
• Extract figures, tables and equations from back/annex sections #1215
• Extract URLs from PDF annotations in fulltext #1315
• Mark consolidated bibliographical references and header explicitly in TEI output #1313
• Include middle name and format initials in BibTeX output #1356
• Fetch ORCID from Crossref when not extracted by Grobid #1406
• Timeout configuration for consolidation requests (separate glutton and Crossref timeout) #1340
• Lingua as an alternative for language recognition #1239
• Blingfire as an alternative sentence segmentation engine #1378
• Native support for Linux ARM 64 architecture
• Multi-architecture Docker builds with ARM64 support (pdfalto and wapiti binaries for Linux ARM 64)
• Support for Python environment managers (virtualenv, conda) for DeLFT integration #1010
• Added version and revision information in the web UI #1390
• Added health status indicator with periodic updates in the web UI #1403
• Added more explanation and links to documentation in the web UI #1391
• More informative /api/health endpoint, failing early when models are partially initialised #1373
• -modelPath CLI argument for training and eval-mode model loading #1383, #1389
• Evaluation script for running end-to-end evaluation from the repository root
• Enabled trivy security code scanning #1295
• Updated Citation.cff and SWID metadata #1341

Changed

• Revised and updated the Crossref integration, with better handling of API limits and errors, in collaboration with Crossref team #1398
• Upgraded to JDK 21 and Gradle 9 #1321
• Updated TensorFlow to 2.17 with Python 3.10-3.11 support #1188
• Updated pdfalto to 0.6.0
• Updated wapiti to 1.5.1
• Updated JEP to 4.2.2 #1332
• Updated DeLFT to > 0.4.1 in documentation and Dockerfiles #1400
• Updated JRuby to 9.4.12.1 and pragmatic segmenter #1293
• Updated Docker base images from deprecated openjdk to eclipse-temurin (21.0.10_7)
• Updated Dropwizard to address Trivy vulnerability in Docker image
• Updated grobid-lucene-analyzers #1346
• Updated dependency versions in build.gradle #1377
• Extensive model retraining: header, segmentation, fulltext, article-light, and article-light-ref models updated across CRF, BidLSTM_CRF_FEATURES, and BidLSTM_ChainCRF_FEATURES architectures
• Significant expansion of training data for segmentation, fulltext, header, name, and affiliation-address models
• Refactored training framework for clearer extensibility #1393
• Updated benchmark results #1392
• Removed obsolete and unused models #1367
• Enhanced documentation structure and clarity for newcomers #1310, #1382
• Return XML by default when no HTTP Accept header is provided #1405
• CI speed-up #1374

Fixed

• Figures, tables and equations identifier uniqueness and overlapping IDs in body and annex #1342
• IndexOutOfBoundException in ORCID search by annotation #1369
• Missing logic to correctly get conflicts and credits in the output TEI
• BibTeX index bug #1409
• Revision link format in the web UI #1404
• German wordforms failing to load in the Lexicon #1362
• Honour instance-level Wapiti params in train() #1383
• Evaluation script now works from the repository root
• Docker build crash caused by dynamic Python environment version fetching #1348
• Dockerfile for ARM Linux #1395
• Full Docker image build restored #1371
• preload_embeddings.py crash when download directory doesn't exist
• Security-oriented regex improvements #1366
• Coveralls build and Gradle deprecations #1347
• Numerous training data corrections across all models

New Contributors

• @jgoodall made their first contribution in #1261
• @haydn-jones made their first contribution in #1301
• @seang096 made their first contribution in #1335
• @homfunc made their first contribution in #1321
• @thesanogoeffect made their first contribution in #1348

Full Changelog: 0.8.2...0.9.0