这里是差集:Docling 全量解析 text block 减去最终会进入正文的 block。优先人工检查 risk=high / medium。
layout_review.html excluded_blocks.json excluded_blocks.tsv final_body_blocks.tsv visual_assets.tsv original.pdf
{
"parsed_text_blocks": 332,
"final_body_blocks": 70,
"excluded_blocks": {
"count": 262,
"by_reason": {
"after_back_matter_stop": 112,
"inside_back_matter": 40,
"docling_page_header": 28,
"outside_body_flow_caption": 16,
"body_heading": 15,
"docling_page_footer": 14,
"inside_front_matter": 12,
"first_page_metadata": 6,
"non_body_heading": 3,
"outside_body_flow": 3,
"front_matter_heading": 2,
"outside_body_flow_list_item": 2,
"page_margin_header": 2,
"abstract_heading": 1,
"back_matter_heading": 1,
"before_body_started": 1,
"early_back_matter_heading": 1,
"front_matter_author_line": 1,
"inside_abstract": 1,
"outside_body_flow_footnote": 1
},
"by_role_guess": {
"reference": 140,
"page_header": 32,
"page_footer": 19,
"caption": 16,
"body_heading": 15,
"body_candidate_excluded": 8,
"unknown_text": 8,
"affiliation": 7,
"back_matter_heading": 3,
"front_matter_heading": 3,
"title_candidate": 3,
"footnote": 2,
"metadata": 2,
"page_margin_header": 2,
"abstract_candidate": 1,
"abstract_heading": 1
},
"by_risk_level": {
"low": 247,
"medium": 8,
"high": 7
},
"high_risk_count": 7,
"medium_risk_count": 8
},
"char_counts": {
"parsed_text_chars": 82901,
"final_body_chars": 49788,
"excluded_chars": 33113
}
}
{
"truncated": true,
"message": "Body extraction stopped at page 16 block #/texts/163: Declaration of Competing Interest. 112 following text blocks were excluded as after_back_matter_stop.",
"stop_trigger": {
"ref": "#/texts/163",
"page": 16,
"layout_order": 219,
"role_guess": "back_matter_heading",
"body_decision_reason": "back_matter_heading",
"text_preview": "Declaration of Competing Interest"
},
"first_truncated_block": {
"ref": "#/texts/206",
"page": 16,
"layout_order": 220,
"role_guess": "reference",
"body_decision_reason": "after_back_matter_stop",
"text_preview": "K.A. Jarvis, C.-C. Wang, A. Manthiram, P.J. Ferreira, J. Mater. Chem. A 2 (2014) 1353-1362."
},
"truncated_block_count": 112,
"truncated_pages": [
16,
17,
18
],
"by_role_guess": {
"reference": 100,
"page_header": 4,
"page_footer": 3,
"affiliation": 2,
"unknown_text": 2,
"body_candidate_excluded": 1
}
}
这些框显示被排除块,同时叠加真实图表资产框。青色虚线表示该 text block 被图表资产 caption 吸收;红色 STOP 是截断触发点,红色框是截断后被排除的块。