feat(metadata): enhance document processing to include metadata extraction and handling

- Updated DocReaderServicer to pass metadata in responses.
- Modified PipelineParser to accumulate and merge metadata from all parsers.
- Enhanced StdWebParser to extract and log the title from web page content.
- Implemented logic in knowledge service to update knowledge title based on extracted metadata.
This commit is contained in:
wizardchen
2026-04-01 14:35:31 +08:00
committed by lyingbug
parent ce7df624fb
commit c4f5db7e88
4 changed files with 32 additions and 4 deletions
+1
View File
@@ -151,6 +151,7 @@ class DocReaderServicer(docreader_pb2_grpc.DocReaderServicer):
markdown_content=_c(result.content),
image_refs=image_refs,
image_dir_path=image_dir,
metadata={k: _c(str(v)) for k, v in result.metadata.items()} if result.metadata else {},
)
logger.info(
"Read response: content_len=%d, images=%d",
+6 -3
View File
@@ -132,8 +132,9 @@ class PipelineParser(BaseParser):
Document: Final document after processing through all parsers,
with accumulated images from all stages
"""
# Accumulate images from all parsers
# Accumulate images and metadata from all parsers
images: Dict[str, str] = {}
metadata: Dict = {}
document = Document()
for p in self._parsers:
logger.info(f"PipelineParser: using parser {p.__class__.__name__}")
@@ -141,10 +142,12 @@ class PipelineParser(BaseParser):
document = p.parse_into_text(content)
# Convert document content back to bytes for next parser
content = endecode.encode_bytes(document.content)
# Accumulate images from this parser
# Accumulate images and metadata from this parser
images.update(document.images)
# Merge all accumulated images into final document
metadata.update(document.metadata)
# Merge all accumulated images and metadata into final document
document.images.update(images)
document.metadata.update(metadata)
return document
@classmethod
+13 -1
View File
@@ -1,5 +1,6 @@
import asyncio
import logging
import re
from playwright.async_api import async_playwright
from trafilatura import extract
@@ -110,7 +111,18 @@ class StdWebParser(BaseParser):
if not md_text:
logger.error("Failed to parse web page")
return Document(content=f"Error parsing web page: {url}")
return Document(content=md_text)
# Extract title from trafilatura metadata output (e.g. "title: xxx" line)
metadata = {}
title_match = re.search(r"^title:\s*(.+)", md_text, re.MULTILINE)
if title_match:
extracted_title = title_match.group(1).strip()
if extracted_title:
metadata["title"] = extracted_title
logger.info(f"Extracted article title from trafilatura: {extracted_title}")
else:
logger.info(f"No title found in trafilatura output, first 200 chars: {md_text[:200]!r}")
return Document(content=md_text, metadata=metadata)
class WebParser(PipelineParser):
+12
View File
@@ -7529,6 +7529,18 @@ func (s *knowledgeService) ProcessDocument(ctx context.Context, t *asynq.Task) e
if convertResult == nil {
return nil
}
// Update knowledge title from extracted page title if not already set
if knowledge.Title == "" || knowledge.Title == payload.URL {
if extractedTitle := convertResult.Metadata["title"]; extractedTitle != "" {
knowledge.Title = extractedTitle
knowledge.UpdatedAt = time.Now()
if err := s.repo.UpdateKnowledge(ctx, knowledge); err != nil {
logger.Warnf(ctx, "Failed to update knowledge title from extracted page title: %v", err)
} else {
logger.Infof(ctx, "Updated knowledge title to extracted page title: %s", extractedTitle)
}
}
}
} else if len(payload.Passages) > 0 {
// Text passage import - direct chunking, no conversion needed
passageChunks := make([]types.ParsedChunk, 0, len(payload.Passages))