v2.3.2

• Improve merging multi-page tables in PdfTabbyReader.
• Stop parsing after client disconnection (for API usage, see issue 488).

v2.3.1

Release note: v2.3.1

• Fix bug with bold lines in DocxReader (see issue 479).
• Upgraded requirements.txt (beautifulsoup4 to 4.12.3 version).
• Added support for external grobid (added support parameter Authorization).
• Added GOST (Russian government standard) frame recognition in PdfTabbyReader (need_gost_frame_analysis parameter).
• Update documentation (added GOST frame recognition).
• Added multi-page table handling to PdfTabbyReader.

v2.3

• Dedoc telegram chat created.
• Added patterns parameter for configuring default structure type.
• Added notebooks with Dedoc usage (see issue 484).
• Fix bug OutOfMemoryError: Java heap space in PdfTabbyReader (see issue 489).
• Fix bug with numeration in DocxReader (see issue 494).
• Added GOST (Russian government standard) frame recognition in PdfImageReader and PdfTxtlayerReader (need_gost_frame_analysis parameter).

v2.2.7

• Fix bugs with start, end of BBoxAnnotation in PdfTabbyReader.
• Improve columns classification and orientation detection for PDF and images (is_one_column_document and document_orientation parameters).
• Upgrade docker: docker-compose is no longer supported, use docker compose instead.
• Fix bug of tables parsing in DocxReader (see issue).
• Added simple textual layer detection in PdfAutoReader (fast_textual_layer_detection parameter).
• Improve paragraph extraction from PDF documents and images.
• Retrain a classifier for diplomas (document_type="diploma") on a new dataset.

v2.2.6

• Upgrade dependencies: numpy<2.0 and dedoc-utils==0.3.7.

v2.2.5

• Added internal functions and classes to support integration of Dedoc into langchain
• Upgrade some dependencies, in particular, xgboost>=1.6.0, pandas, pdfminer.six

v2.2.4

• Show page division and page numbers in the HTML output representation (API usage, return_format="html").
• Make imports from dedoc library faster.
• Added tutorial how to add a new language to dedoc (not finished entirely).
• Added additional page_id metadata for multi-page nodes (structure_type="tree" in API, TreeConstructor in the library).
• Updated OCR and orientation/columns classification benchmarks.
• Minor edits of README.md.
• Fixed empty cells handling in CSVReader.
• Fixed bounding boxes extraction for text in tables for PdfTabbyReader.

v2.2.3

• Show attached images and added ability to download attached files in the HTML output representation (API usage, return_format="html").
• Added hierarchy level information and annotations to PptxReader.

v2.2.2

• Added images extraction to ArticleReader.
• Added attachments and references to them in the HTML output representation (return_format="html").
• Fixed functionality of parameter need_content_analysis.
• Fixed CSVReader (exclude BOM character from the output).
• Added handling files with wrong extension or without extension to DedocManager (detect file type by its content).
• Update README.md.

v2.2.1

• Added fintoc structure type for parsing financial prospects according to the FinTOC 2022 Shared task (FintocStructureExtractor).
• Fixed small bugs in ArticleReader: colspan for tables, keywords, sections numbering, etc.
• Added references to nodes and fixed small bugs in the HTML output representation (return_format="html").
• Removed other_fields from LineMetadata and DocumentMetadata.
• Update README.md.