Enhance table title handling and improve heading detection logic

- Added logic to set `table_title` from `caption` if it is under 160 characters and not already set.
- Updated `_blocks_from_content_list` to assign `table_title` based on `caption` length.
- Introduced new regex patterns for better detection of TOC entries and noise.
- Enhanced heading detection to differentiate between numbered instructions and actual headings.
- Added tests to verify that table captions are correctly assigned as titles and that numbered instructions are treated as body text.

backend/rag_cut/splitters/heading_splitter.py

@@ -7,9 +7,27 @@ from dataclasses import dataclass, field
 
 from rag_cut.models import Block, BlockType, SplitConfig
 
-# e.g. "1.2 ACCOUNT STATUS CODE MASTER", "10.上升三角形態" (space after '.' optional)
+# e.g. "1.2 ACCOUNT STATUS CODE MASTER", "10.上升三角形態".
+# The title starts with a letter/CJK character so "1.1.1" cannot backtrack
+# into number="1" + title="1.1".
 NUMBERED_HEADING_RE = re.compile(
-    r"^\s*(\d+(?:\.\d+)*)(?:\.|．)?\s*([A-Za-z0-9\u4e00-\u9fff][A-Za-z0-9\u4e00-\u9fff&/ \-_:：]{2,})\s*$"
+    r"^\s*(\d+(?:\.\d+)*)(?:[.．、:：]\s*|\s+)"
+    r"([A-Za-z\u4e00-\u9fff][^\n]{1,120})\s*$"
+)
+PURE_NUMBERED_HEADING_RE = re.compile(r"^\s*(\d+(?:\.\d+){1,5})\.?\s*$")
+TIME_LIKE_RE = re.compile(r"^\s*\d{1,2}:\d{2}(?::\d{2})?\s*$")
+NUMBERED_LIST_ITEM_RE = re.compile(r"^\s*\d+[.)．、]\s+(?P<body>.+)$")
+NUMBERED_OPTION_SENTENCE_RE = re.compile(
+    r"^\s*\d+(?:\.\d+)+\.?\s+.+\bthis\s+(?:option|function|feature)\s+"
+    r"(?:enables|allows)\b",
+    re.I,
+)
+INSTRUCTION_START_RE = re.compile(
+    r"^(?:select|click|choose|enter|input|open|close|press|perform|to\s+|on\s+|"
+    r"the\s+user|users?\s+|next\s+|then\s+|for\s+ease|option/tool|"
+    r"用户|点击|輸入|输入|選擇|选择|填写|當|当|在|首先|然后|然後|配置|系统|系統|"
+    r"若|如果|注|详细|詳細|列表|机构|機構|添加|删除|刪除)",
+    re.I,
 )
 FIGURE_TABLE_RE = re.compile(
     r"^\s*(?:Figure|Fig\.|图|表|Table)\s*[\d.]+",
@@ -38,12 +56,39 @@ def _rendered_len(blocks: list[Block]) -> int:
 
 
 def numbered_heading_level(text: str) -> int | None:
-    match = NUMBERED_HEADING_RE.match(text.strip())
+    stripped = text.strip()
+    match = PURE_NUMBERED_HEADING_RE.match(stripped)
+    if match:
+        return match.group(1).count(".") + 1
+    match = NUMBERED_HEADING_RE.match(stripped)
     if not match:
         return None
     return match.group(1).count(".") + 1
 
 
+def looks_like_numbered_instruction(text: str) -> bool:
+    """True for numbered procedure/list sentences, not outline headings."""
+    stripped = text.strip()
+    if NUMBERED_OPTION_SENTENCE_RE.match(stripped):
+        return True
+    match = NUMBERED_LIST_ITEM_RE.match(stripped)
+    if not match:
+        return False
+    body = match.group("body").strip()
+    if INSTRUCTION_START_RE.match(body):
+        return True
+    return len(body) >= 45 and bool(re.search(r"[,.，。;；:]", body))
+
+
+def looks_like_false_heading_text(text: str) -> bool:
+    stripped = text.strip()
+    if TIME_LIKE_RE.match(stripped) or looks_like_numbered_instruction(stripped):
+        return True
+    if numbered_heading_level(stripped):
+        return False
+    return len(stripped) >= 40 and bool(re.search(r"[.!?。！？；;]\s*$", stripped))
+
+
 def infer_heading_level(block: Block) -> int:
     if block.type == BlockType.HEADING and block.level:
         numbered = numbered_heading_level(block.text)
@@ -60,6 +105,8 @@ def is_heading_block(block: Block) -> bool:
     text = _text(block)
     if not text or len(text) > MAX_HEADING_CHARS:
         return False
+    if looks_like_false_heading_text(text):
+        return False
     if block.type == BlockType.HEADING:
         return True
     if NUMBERED_HEADING_RE.match(text):
@@ -71,14 +118,21 @@ def is_heading_block(block: Block) -> bool:
 
 def normalize_heading_block(block: Block) -> Block:
     text = _text(block)
-    if block.type == BlockType.HEADING and len(text) > MAX_HEADING_CHARS:
+    false_heading = looks_like_false_heading_text(text)
+    if block.type == BlockType.HEADING and (
+        len(text) > MAX_HEADING_CHARS or false_heading
+    ):
         # Parser sometimes merges title+body then marks the blob as heading.
         return Block(type=BlockType.PARAGRAPH, text=text, meta=dict(block.meta))
+    if false_heading:
+        return block
     numbered = numbered_heading_level(text)
     if block.type == BlockType.HEADING:
         level = numbered or block.level or 1
         return block.model_copy(update={"level": level})
-    if numbered and NUMBERED_HEADING_RE.match(text):
+    if numbered and (
+        NUMBERED_HEADING_RE.match(text) or PURE_NUMBERED_HEADING_RE.match(text)
+    ):
         return Block(
             type=BlockType.HEADING,
             text=text,