| """Render-faithful extraction for Kiwix MediaWiki HTML (candidate v2). |
| |
| Fixes measured on 2026-09-14 against the current extractor: |
| 1. per-article ZIM footer boilerplate ("This article is issued from Wikipedia…") |
| present in 96.1% of published ja chunks / 89.9% of zh chunks; |
| 2. <head>/<h1> title text duplicated into the body; |
| 3. inline tags replaced by a space, splitting words ("<b>Ar</b>gentino" -> "Ar gentino"); |
| 4. HTML entities never unescaped ( , & survive as text); |
| 5. injected spaces inside CJK sentences. |
| Block elements become newlines so the chunker can treat them as segment boundaries. |
| """ |
| import re |
| from html import unescape |
|
|
| _CONTENT = re.compile(r'<div[^>]+id="mw-content-text"', re.I) |
| _HTDIG = re.compile(r'<!--\s*htdig_noindex\s*-->.*?<!--\s*/htdig_noindex\s*-->', re.S | re.I) |
| _ZIMFOOT = re.compile(r'<div[^>]*class="[^"]*zim-footer[^"]*".*?</div>', re.S | re.I) |
| _SCRIPT = re.compile(r'<(script|style)\b[^>]*>.*?</\1>', re.I | re.S) |
| _COMMENT = re.compile(r'<!--.*?-->', re.S) |
| _BLOCK = re.compile(r'</?(?:p|div|section|h[1-6]|li|ul|ol|dl|dd|dt|tr|table|tbody|thead|br|hr|blockquote|figure|figcaption|caption|aside|nav|header|footer|main)\b[^>]*>', re.I) |
| _CELL = re.compile(r'</(?:td|th)\s*>', re.I) |
| _TAG = re.compile(r'<[^>]+>') |
| _CSSCMT = re.compile(r'/\*.*?\*/', re.S) |
| _CSSRULE = re.compile(r'[.#@][-\w][^{}<>]{0,200}\{[^{}]*\}') |
| _FOOTTXT = re.compile(r'This article is issued from Wikipedia[^\n]*', re.I) |
| _EDITLINKS = re.compile(r'\[\s*(?:edit|編集|编辑|編輯)\s*\]', re.I) |
| _NAVLINE = re.compile(r'^(?:[■□▪\s]*(?:Template|テンプレート|ノート|解説|Project|プロジェクト|カテゴリ|表示|編集|v\s*[·・.]?\s*t\s*[·・.]?\s*e|查|论|編|编|論|閱|读)\b[^\n]{0,40})$', re.I) |
| _SYMBOLONLY= re.compile(r'^[\W_]{1,12}$') |
| _CJK = r' -〿-ヿ㐀-䶿一-鿿豈--' |
| _CJKSPACE = re.compile(r'(?<=[' + _CJK + r'])[ \t]+(?=[' + _CJK + r'])') |
| _SPACES = re.compile(r'[ \t ]+') |
| _NEWLINES = re.compile(r'\s*\n\s*') |
|
|
| def extract_text(html: str) -> str: |
| m = _CONTENT.search(html) |
| if m: |
| html = html[m.start():] |
| else: |
| b = html.lower().find('<body') |
| if b >= 0: html = html[b:] |
| html = _HTDIG.sub('\n', html) |
| html = _ZIMFOOT.sub('\n', html) |
| html = _SCRIPT.sub(' ', html) |
| html = _COMMENT.sub(' ', html) |
| html = _CELL.sub(' | ', html) |
| html = _BLOCK.sub('\n', html) |
| html = _TAG.sub('', html) |
| text = unescape(html) |
| text = _CSSCMT.sub(' ', text) |
| text = _CSSRULE.sub(' ', text) |
| text = _FOOTTXT.sub(' ', text) |
| text = _EDITLINKS.sub(' ', text) |
| text = text.replace(' | \n', '\n').replace(' |', ' |') |
| text = _SPACES.sub(' ', text) |
| text = _NEWLINES.sub('\n', text) |
| text = _CJKSPACE.sub('', text) |
| lines, seen_first = [], None |
| for ln in (l.strip(' |') for l in text.split('\n')): |
| if not ln: continue |
| if _NAVLINE.match(ln) or _SYMBOLONLY.match(ln): continue |
| if seen_first is None: seen_first = ln |
| elif ln == seen_first: continue |
| lines.append(ln) |
| return '\n'.join(lines).strip() |
|
|
