Repository navigation
Open
Labels
P2Priority 2: important, scheduled after P0/P1Priority 2: important, scheduled after P0/P1enhancementNew feature or requestNew feature or request
Description
Summary
Add support for extracting review comments from Word documents across both modern OOXML formats (.docx, .docm, .dotx) and legacy Word 97–2003 binary OLE documents (.doc).
User Value
Users and customers frequently store critical review feedback, clarifications, and Threaded Q&A in Word margin comments. Previously:
.docxfiles processed by Azure Document Intelligence (prebuilt-read) only extracted main document text, headers/footers, tables, and footnotes, ignoringword/comments.xml..docmfiles processed viadocx2txtignoredword/comments.xml..doc(Word 97–2003 OLE) files flattened the piece table without extracting structured comment metadata (ccpAtn,PlcfandTxt,PlcfandRef,GrpXstAtnOwners), author names, or anchor positions.
Extracting and indexing Word comments alongside the document chunks where they are anchored makes review comments searchable and available to AI chat, workspace RAG, M365 extraction, and content screening.
Proposed Behavior
- OOXML Word files (
.docx,.docm,.dotx):- Safely parse
word/comments.xml(author, initials, date, comment text),word/commentsExtended.xml(threaded replies viaparaIdParentand resolved/done state), andword/document.xml(w:commentRangeStart,w:commentRangeEnd,w:commentReferencefor quoted anchor text and reading-orderword_offset).
- Safely parse
- Legacy binary Word files (
.doc):- Parse the MS-DOC File Information Block (
FibRgLw97ccpAtnandFibRgFcLcb97PlcfandTxt,PlcfandRef,GrpXstAtnOwners) to extract structured comments with author/initials and anchor positions while keeping raw annotation subdocument text separate from body text.
- Parse the MS-DOC File Information Block (
- Chunk & Text Integration:
- Merge extracted comments into the chunk containing their anchored text during workspace ingestion (
process_di_document,process_office_template), include them inextract_word_text()/ direct chat uploads, and pass them to content screening.
- Merge extracted comments into the chunk containing their anchored text during workspace ingestion (
Acceptance Criteria
-
.docx,.docm, and.dotxcomments (including author, date, anchored text, threaded replies, and resolved status) are extracted and indexed. - Legacy
.docOLE comments are extracted with author/initials and anchor context without duplicating raw annotation text in the body. - Extracted comments in
.docxdocuments are merged into their corresponding reading-order chunk (with overflow fallback). - Functional tests cover both OOXML
.docxand legacy.doccomment extraction and chunk placement, and release notes are updated.
Notes
- Uses local bounded XML (
defusedxml) and OLE (olefile) parsing with zero additional Azure API calls.
Metadata
Metadata
Assignees
Labels
P2Priority 2: important, scheduled after P0/P1Priority 2: important, scheduled after P0/P1enhancementNew feature or requestNew feature or request
Type
Projects
- StatusShow more project fieldsIn progress
Activity