webpage-parser 0.1.0__py3-none-any.whl

webpage_parser/__init__.py

@@ -0,0 +1,11 @@
+"""网页解析算子：从爬取到的 HTML 抽取 title/content/pub_time/author。
+
+包内模块沿用「相对导入优先、扁平导入兜底」的双模式写法，因此既可以按包导入
+（`from webpage_parser import process_row`），也可以把目录直接摊平当脚本跑
+（`python3 extract.py`）。推荐按包导入。
+"""
+__version__ = "0.1.0"
+
+from .pipeline import empty_result, process_row
+
+__all__ = ["empty_result", "process_row", "__version__"]

webpage_parser/author.py

@@ -0,0 +1,503 @@
+"""author 抽取：候选枚举 + 强拒绝（V3 以否定式条款为主）。
+
+候选链：JSON-LD author -> meta author 键 -> extra.anchor（AuthorList 系且可见文本命中）
+        -> 可见文本 byline 正则。
+强拒绝：站点名 / 编辑角色 / 占位符 / @handle / 邮箱 / URL。
+"""
+import re
+
+try:
+    from .util import clean_text, norm_key, has_cjk
+    from .htmlprep import ARTICLE_TYPES, walk_json_keys
+    from .title import _site_names, _confirmed_site
+except ImportError:
+    from util import clean_text, norm_key, has_cjk
+    from htmlprep import ARTICLE_TYPES, walk_json_keys
+    from title import _site_names, _confirmed_site
+
+_AUTHOR_META_KEYS = ("author", "article:author", "byline", "byl", "dc.creator",
+                     "og:author", "sogou_author", "og:article:author",
+                     "sailthru.author", "parsely-author", "mrf:authors",
+                     "authors", "dable:author", "cxenseparse:author",
+                     "articleauthor", "dcterms.creator", "creator",
+                     "mediaid")  # mediaid：搜狐号署名 meta（验收反馈 case_08）
+# twitter:creator 刻意排除：实测值全是 @站号句柄（调研05 §6.3）
+# 标签剥离：多字标签冒号可选；单字「文」必须带冒号，否则会把「文娱阁」这类
+# 以文开头的署名截掉首字（验收反馈 case_09：anchor/页面署名均为「文娱阁」被截成「娱阁」）
+_LABEL_ACCEPT_RE = re.compile(
+    r"^(?:(?:\u4f5c\u8005|\u8bb0\u8005|\u64b0\u7a3f|\u64b0\u6587|\u4f5c\u8005\u540d|"
+    r"\u7279\u7ea6\u4f5c\u8005|\u901a\u8baf\u5458|by|author)\s*[:：]?\s*|\u6587\s*[:：]\s*)",
+    re.IGNORECASE,
+)
+_LABEL_REJECT_RE = re.compile(
+    r"^(?:\u7f16\u8f91|\u8d23\u7f16|\u8d23\u4efb\u7f16\u8f91|\u5ba1\u6838|\u6821\u5bf9|"
+    r"\u503c\u73ed\u7f16\u8f91|\u6765\u6e90|\u51fa\u5904|\u6587\u7ae0\u6765\u6e90|\u7a3f\u6e90|"
+    r"\u7533\u660e|\u58f0\u660e)\s*[:：]?",
+    re.IGNORECASE,
+)
+_PLACEHOLDER_RE = re.compile(r"\{\{|%\w+%|\$\{|<\{|unknown|anonymous", re.IGNORECASE)
+# 候选含域名 → 整串是站点口号/署名垃圾，拆分前整体拒绝（dm5 meta author 实测）
+_DOMAIN_IN_NAME_RE = re.compile(
+    r"\b[\w-]+\.(?:com|net|org|cn|io|co|cc|tv|me|xyz|top|vip|site|info)\b",
+    re.IGNORECASE)
+# 纯英文角色词整串：Assistant Editor / Managing Editor 等是角色不是姓名
+_EN_ROLE_RE = re.compile(
+    r"^(?:(?:assistant|associate|managing|senior|deputy|contributing|executive|"
+    r"chief|staff|web|news|copy|digital|online|commissioning)\s+)*"
+    r"(?:editor|writer|correspondent|reporter|columnist|producer|publisher)s?\.?$",
+    re.IGNORECASE)
+_EMAIL_RE = re.compile(r"[\w.+-]+@[\w-]+\.[\w.]+")
+_URL_RE = re.compile(r"https?://|www\.", re.IGNORECASE)
+_HANDLE_RE = re.compile(r"^@[\w\u4e00-\u9fff]+$")
+_ROLE_TAIL_RE = re.compile(r"(\u7f16\u8f91|\u8d23\u7f16|\u5ba1\u6838|\u6821\u5bf9)$")
+_USERNAME_RE = re.compile(
+    r"^(admin|administrator|user\d*|guest\d*|test\d*|editor\d*|"
+    r"\u6e38\u5ba2|\u697c\u4e3b|\u533f\u540d\u7528\u6237)$", re.IGNORECASE)
+_COPYRIGHT_RE = re.compile(r"\u7248\u6743|\u00a9|copyright", re.IGNORECASE)
+_UI_NOISE_RE = re.compile(
+    r"\u8bc4\u8bba|\u56de\u590d|\u9605\u8bfb|\u70b9\u8d5e|\u5206\u4eab|\u5173\u6ce8|"
+    r"\u8ba2\u9605|\u6536\u85cf|\u8f6c\u53d1|\u767b\u5f55|\u6ce8\u518c|\u8bbf\u95ee|"
+    r"share|comment|subscribe|sign\s*in|log\s*in|preferred\s+on\s+google|add\s+as\s+preferred|\u4ec5\u4ee3\u8868\u4f5c\u8005|\u67e5\u770b\u4f5c\u8005\u7b80\u4ecb", re.IGNORECASE)
+_DATE_FRAG_RE = re.compile(
+    r"(?:19|20)\d{2}[-/\u5e74.]\s*\d{1,2}[-/\u6708.]\s*\d{1,2}\s*\u65e5?"
+    r"(?:\s*\d{1,2}[:：\u65f6]\d{2}(?:[:：\u5206]\d{2}\u79d2?)?)?"
+    r"|\d{4}-\d{2}-\d{2}[Tt ]\d{2}:\d{2}(:\d{2})?"
+    r"|(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)[a-z]*\.?\s+\d{1,2}(?:st|nd|rd|th)?(?:,?\s*\d{4})?"
+    r"|\d{1,2}\s+(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)[a-z]*\.?(?:\s*\d{4})?"
+    r"|\d+\s*(?:seconds?|minutes?|hours?|days?|weeks?|months?)\s+ago"
+    r"|a\s+(?:day|week|month|year|few\s+(?:days|hours|minutes))\s+ago"
+    r"|just\s+now"
+    r"|\b\d{1,2}:\d{2}\s*(?:a\.?m\.?|p\.?m\.?)\b(?:\s*[A-Z]{2,4})?"
+    r"|\d+\s*\u4e2a?\s*(?:\u79d2|\u5206\u949f|\u5c0f\u65f6|\u5929|\u65e5|\u5468|\u661f\u671f)\u524d"
+    r"|\u6628\u5929|\u4eca\u5929|\u524d\u5929|\u521a\u521a"
+    r"|(?:published|updated|posted|added|by)\b",
+    re.IGNORECASE)
+_CN_TAIL_SIGN_RE = re.compile(
+    r"[\uff08(]\s*(?:[\u4e00-\u9fff]{0,6}?\u8bb0\u8005)\s*"
+    r"([^\uff08\uff09()\uff0c\uff0c\u3002\uff1b;,]{2,20}?)\s*[\uff09)]"
+    r"|\u6587\s*[\uff5c|/]\s*([^\s\uff0c\u3002,;]{2,15})"
+)
+_FOLLOW_TAIL_RE = re.compile(
+    r"(\u5173\u6ce8|\u8ba2\u9605|\u52a0\u5173\u6ce8|follow|subscribe)$", re.IGNORECASE)
+_BYLINE_RE = re.compile(
+    r"(?:\u4f5c\u8005|\u8bb0\u8005|\u6587|\u64b0\u7a3f|\u64b0\u6587)\s*[:：]\s*"
+    r"([^\s\uff0c\u3002\uff1b\uff1a|,;:(\uff08)\uff09\u201c\u201d\u2018\u2019\"\x27+]{2,20})"
+)
+_BYLINE_EN_RE = re.compile(r"\bBy\s+([A-Z][A-Za-z.\-]+(?:\s+[A-Z][A-Za-z.\-]+){0,3})")
+_WIRE_BYLINE_RE = re.compile(
+    r"(?<![A-Za-z,])([A-Z][A-Za-z.'\-]+(?:\s+[A-Z][A-Za-z.'\-]+){1,3}),\s+"
+    r"(?:The\s+)?[A-Z][A-Za-z.&' ]{1,40}?"
+    r"(?:Press|News|Media|Wire|Times|Post|Herald|Tribune|Reuters|AFP|AP)\.?(?:\s|$)")
+_WIRE_ORG_IN_NAME_RE = re.compile(
+    r"\b(Press|News|Media|Wire|Times|Post|Herald|Tribune|Reuters|AFP|AP)\b")
+
+
+def _wire_byline_candidates(tail):
+    pos = 0
+    while pos < len(tail):
+        m = _WIRE_BYLINE_RE.search(tail, pos)
+        if not m:
+            return
+        name = m.group(1)
+        if _WIRE_ORG_IN_NAME_RE.search(name):
+            pos = m.start() + 1
+            continue
+        yield name
+        pos = m.end()
+_TITLE_SEG_RE = re.compile(r"[_\-|｜–—»]+")
+_SECTION_WORDS = {
+    "焦点", "时尚", "娱乐", "体育", "财经", "科技", "军事", "国际", "国内",
+    "社会", "专题", "汽车", "房产", "游戏", "教育", "健康", "核心提示", "要闻",
+    "租借",
+}
+_ORG_HINT_RE = re.compile(
+    r"\u793e|\u7f51|\u62a5|\u53f0|\u5a92\u4f53|\u4e2d\u5fc3|\u90e8|\u5c40|\u59d4\u5458\u4f1a|"
+    r"\u516c\u53f8|\u96c6\u56e2|\u5927\u5b66|\u7814\u7a76\u9662|news|media|agency|studio$",
+    re.IGNORECASE,
+)
+_SPLIT_RE = re.compile(r"[\u3001\uff0c,|/;]+|\s{2,}")
+_MAX_AUTHORS = 8
+
+
+def _norm_cmp(s):
+    return norm_key(s)
+
+
+def _reject(name, site_cfg, title_norm, site_names):
+    n = clean_text(name)
+    if not n:
+        return True
+    nn = _norm_cmp(n)
+    if not nn:
+        return True
+    if len(n) > 60:
+        return True
+    if n.isdigit():
+        return True
+    if "=" in n or "&" in n:
+        return True
+    if n in _SECTION_WORDS:
+        return True
+    if re.match(r"^\d", n):
+        return True
+    if re.match(r"^\u6765\u81ea", n):
+        return True
+    if re.match(r"^\u5173\u4e8e", n):
+        return True
+    if re.search(r"\uae30\uc790|\uae30\uc0ac|\ubaa8\uc74c|\uc785\ub825", n) and len(n) <= 6:
+        return True
+    if _PLACEHOLDER_RE.search(n):
+        return True
+    if _EMAIL_RE.search(n) or _URL_RE.search(n) or _HANDLE_RE.match(n):
+        return True
+    if _USERNAME_RE.match(n):
+        return True
+    if _COPYRIGHT_RE.search(n):
+        return True
+    if _UI_NOISE_RE.search(n):
+        return True
+    if re.search(r"\bstaff\b", n, re.IGNORECASE):
+        return True
+    if _LABEL_REJECT_RE.match(n):
+        return True
+    if _ROLE_TAIL_RE.search(n) and len(n) <= 6:
+        return True
+    if has_cjk(n):
+        if len(n) > 20 or re.search(r"[\u3002\uff01\uff1f]", n):
+            return True
+    else:
+        if len(n.split()) > 8:  # 拉丁整句按词数判定（03 §5.3）
+            return True
+    for bad in site_cfg.reject_names:
+        if nn == _norm_cmp(bad):
+            return True
+    if site_names and _confirmed_site(n, site_names):
+        return True
+    if title_norm and nn == title_norm:
+        return True
+    if nn in ("null", "none", "n/a", "\u4f5a\u540d", "\u533f\u540d", "\u65e0",
+              "\u6682\u65e0", "\u5c0f\u7f16", "\u7efc\u5408", "\u7f51\u7edc",
+              "\u7f51\u53cb", "\u8f6c\u8f7d", "\u5176\u4ed6", "staff", "file",
+              "\u7528\u6237\u540d", "newsprovided",
+              "\u53d1\u5e03\u4e8e", "\u53d1\u8868\u4e8e",
+              "on", "by", "at", "in", "of", "the", "and", "posted"):
+        return True
+    return False
+
+
+def _strip_labels(name):
+    """循环剥标签前缀；若命中拒绝型标签（编辑/来源），返回 None。"""
+    cur = clean_text(name)
+    for _ in range(3):
+        if _LABEL_REJECT_RE.match(cur):
+            return None
+        new = _LABEL_ACCEPT_RE.sub("", cur)
+        if new == cur:
+            break
+        cur = new.strip()
+    return cur
+
+
+def _split_multi(name):
+    if has_cjk(name):
+        parts = [p.strip() for p in re.split(r"[\u3001\uff0c,|/;]+|\s+", name) if p.strip()]
+    else:
+        parts = [p.strip() for p in _SPLIT_RE.split(name) if p.strip()]
+    return parts[:_MAX_AUTHORS] if parts else []
+
+
+def _finalize(cand, site_cfg, title_norm, page_norms, source, site_names):
+    if not cand:
+        return None
+    stripped = _strip_labels(cand)
+    if stripped is None:
+        return None
+    if _URL_RE.search(stripped) or _EMAIL_RE.search(stripped):
+        return None
+    if _DOMAIN_IN_NAME_RE.search(stripped):
+        return None
+    if _EN_ROLE_RE.match(stripped.strip()):
+        return None
+    out = []
+    for p in _split_multi(stripped):
+        if _reject(p, site_cfg, title_norm, site_names):
+            continue
+        if _EN_ROLE_RE.match(p):
+            continue
+        np = _norm_cmp(p)
+        if page_norms and np not in page_norms[0] and np not in page_norms[1]:
+            continue
+        if p not in out:
+            out.append(p)
+    if not out:
+        return None
+    joined = "\u3001".join(out)
+    return joined, source, bool(_ORG_HINT_RE.search(joined))
+
+
+def _jsonld_names(obj, out, depth=0):
+    if depth > 4:
+        return
+    if isinstance(obj, str):
+        if obj.strip():
+            out.append(obj.strip())
+        return
+    if isinstance(obj, dict):
+        nm = obj.get("name")
+        if isinstance(nm, str):
+            out.append(nm.strip())
+            return
+        if isinstance(nm, dict):
+            v = nm.get("@value") or nm.get("name")
+            if isinstance(v, str):
+                out.append(v.strip())
+                return
+        for k in ("author", "creator", "contributor"):
+            if k in obj:
+                _jsonld_names(obj[k], out, depth + 1)
+    elif isinstance(obj, list):
+        for item in obj[:10]:
+            _jsonld_names(item, out, depth + 1)
+
+
+def _visible_tail(signals, limit=4000):
+    """从解析树取可见文本尾部（跳过 script/style，保留词边界）。"""
+    tree = getattr(signals, "tree", None)
+    if tree is None:
+        return ""
+    parts = []
+
+    def walk(el):
+        if not isinstance(el.tag, str):
+            return
+        if el.tag in ("script", "style", "noscript", "template"):
+            return
+        if el.text:
+            parts.append(el.text)
+        for ch in el:
+            walk(ch)
+            if ch.tail:
+                parts.append(ch.tail)
+
+    try:
+        body = tree.find("body")
+        walk(body if body is not None else tree)
+    except Exception:
+        return ""
+    return clean_text(" ".join(parts))[-limit:]
+
+
+def _l1b_author(signals, site_cfg):
+    """L1b 内嵌状态 JSON 的作者名（仅 allow_l1b 站点族）。
+
+    百家号 bsdata...itemdata.name、小红书 note.user.nickname。
+    base 必须在白名单键内且路径含 hint，防止抓到文章标题的 name。
+    """
+    hints = getattr(site_cfg, "l1b_author_hints", ()) or ()
+    if not hints:
+        return ""
+    base_ok = ("name", "nickname", "authorname", "media_name")
+    for obj in signals.state_objs:
+        for key, val in walk_json_keys(obj):
+            base = key.rsplit(".", 1)[-1]
+            if base not in base_ok or not isinstance(val, str):
+                continue
+            if not (2 <= len(val.strip()) <= 40):
+                continue
+            if any(h in key for h in hints):
+                return val.strip()
+    return ""
+
+
+def _clean_byline_node(txt):
+    """署名节点文本清洗：'By Allyson Castillo' / '小周侃球关注' /
+    'John Sicher a day ago' / '大河报·豫视频 2026-08-28 08:15' -> 干净名字或 None。"""
+    if not txt:
+        return None
+    t = clean_text(txt)
+    if not t or len(t) > 60:
+        return None
+    if re.search(r"[。！？]", t):
+        return None
+    t = _EMAIL_RE.sub("", t)
+    t = _DATE_FRAG_RE.sub(" ", t)
+    t = _FOLLOW_TAIL_RE.sub("", t).strip(" ·|｜,，-—–")
+    t = _LABEL_ACCEPT_RE.sub("", t).strip()
+    t = re.sub(r"\s{2,}", " ", t).strip(" ·|｜,，")
+    if not t or len(t) > 40:
+        return None
+    if _DATE_FRAG_RE.search(t):  # 清完还剩日期 -> 本身就不是署名块
+        return None
+    if len(t) <= 12 and _UI_NOISE_RE.search(t):
+        return None
+    if not re.search(r"[A-Za-z\u4e00-\u9fff\uac00-\ud7a5\u3040-\u30ff\u0400-\u04ff\u0600-\u06ff]", t):
+        return None
+    return t
+
+
+_MEDIA_ORG_RE = re.compile(
+    r"(\u65e5\u62a5|\u665a\u62a5|\u65f6\u62a5|\u5feb\u62a5|\u5468\u520a|\u6742\u5fd7|\u65b0\u95fb"
+    r"|\u7535\u89c6\u53f0|\u5e7f\u64ad|\u901a\u8baf\u793e|\u51fa\u7248|\u62a5\u4e1a|\u4f20\u5a92"
+    r"|\u5a92\u4f53|\u5ba2\u6237\u7aef|\u65b0\u6d6a|\u7f51\u6613|\u817e\u8baf|\u51e4\u51f0"
+    r"|sina|sohu|163|qq\.|ifeng|cctv|xinhua|people|chinanews"
+    r"|thepaper)|(\u7f51|\u62a5|\u53f0|\u793e)$", re.I)
+
+
+def _anchor_context_ok(anchor, signals):
+    """anchor 的前文若是「来源：/转载自」且 anchor 本身是媒体机构名，则是内容来源
+    而非作者（03 §5.1b 反例 anchor=新浪财经）；自媒体账号名放行（163 模板里
+    「来源: 王姐懒人家常菜」中的账号名就是网易号作者）。
+
+    返回 True=可以当作者候选。
+    """
+    tree = getattr(signals, "tree", None)
+    if tree is None or not anchor:
+        return False
+    try:
+        body = tree.find("body")
+        root = body if body is not None else tree
+        parts = []
+
+        def walk(el):
+            if not isinstance(el.tag, str):
+                return
+            if el.tag in ("script", "style", "noscript", "template"):
+                return
+            if el.text:
+                parts.append(el.text)
+            for ch in el:
+                walk(ch)
+                if ch.tail:
+                    parts.append(ch.tail)
+
+        walk(root)
+        text = clean_text(" ".join(parts))
+    except Exception:
+        return False
+    idx = text.find(anchor)
+    if idx < 0:
+        return False
+    pre = text[max(0, idx - 8):idx]
+    if re.search(r"(\u6765\u6e90|\u8f6c\u8f7d\u81ea|\u51fa\u5904)\s*[:\uff1a/]?\s*$", pre) \
+            and _MEDIA_ORG_RE.search(anchor):
+        return False
+    return True
+
+
+def _cn_tail_sign(signals):
+    """正文末尾中文记者署名：（记者 张三）/（本报记者 李四）/文｜王五。"""
+    tail = _visible_tail(signals, limit=2500)
+    if not tail:
+        return ""
+    for m in _CN_TAIL_SIGN_RE.finditer(tail):
+        name = m.group(1) or m.group(2) or ""
+        name = re.sub(r"^[\u4e00-\u9fff]{0,8}?\u8bb0\u8005", "", name)
+        name = re.sub(r"\s*\u901a\u8baf\u5458\s*", "\u3001", name).strip()
+        if name:
+            return name
+    return ""
+
+
+def extract_author(signals, site_cfg, extra, visible_norm, html_norm, title=""):
+    """返回 (author, author_source, author_is_org)。"""
+    page_norms = (visible_norm or "", html_norm or "")
+    title_norm = _norm_cmp(title)
+
+    site_names = _site_names(signals, site_cfg)
+    for seg in _TITLE_SEG_RE.split(signals.title_tag or ""):
+        seg = clean_text(seg)
+        if len(seg) >= 2:
+            site_names.add(_norm_cmp(seg))
+
+    # 1) JSON-LD author
+    for node in signals.jsonld_nodes:
+        t = node.get("@type")
+        types = set()
+        if isinstance(t, str):
+            types.add(t.lower())
+        elif isinstance(t, list):
+            types.update(str(x).lower() for x in t)
+        if types and not (types & ARTICLE_TYPES):
+            continue
+        if "author" not in node:
+            continue
+        names = []
+        _jsonld_names(node["author"], names)
+        for nm in names:
+            r = _finalize(nm, site_cfg, title_norm, page_norms, "jsonld", site_names)
+            if r:
+                return r
+
+    # 2) meta author 键
+    for key in _AUTHOR_META_KEYS:
+        for v in signals.meta.get(key, []):
+            r = _finalize(v, site_cfg, title_norm, page_norms, "meta:" + key, site_names)
+            if r:
+                return r
+
+    # 3) extra.anchor（AuthorList 系白名单 + 可见文本命中 + 非来源语境，调研05 §5.1b）
+    anchor = ""
+    try:
+        anchor = clean_text(str((extra or {}).get("anchor") or ""))
+    except Exception:
+        anchor = ""
+    if (anchor and site_cfg.allow_anchor_author
+            and _norm_cmp(anchor) in page_norms[0]
+            and _anchor_context_ok(anchor, signals)):
+        r = _finalize(anchor, site_cfg, title_norm, page_norms, "extra_anchor", site_names)
+        if r:
+            return r
+
+    # 3b) L1b 内嵌状态 JSON 的作者名（百家号 itemdata.name / 小红书 note.user.nickname）
+    if site_cfg.allow_l1b:
+        nm = _l1b_author(signals, site_cfg)
+        if nm:
+            r = _finalize(nm, site_cfg, title_norm, page_norms, "state_json", site_names)
+            if r:
+                return r
+
+    # 4) 署名节点（class/id/testid/itemprop/rel 命中），清洗后过强拒绝
+    node_cands = []
+    for txt, strong_bz, weak_bz, pos, tag in getattr(signals, "author_nodes", []):
+        if strong_bz:
+            continue
+        c = _clean_byline_node(txt)
+        if not c:
+            continue
+        node_cands.append((0 if pos < 0.5 else 1, len(c), c))
+    node_cands.sort()
+    for _score, _ln, cand in node_cands[:6]:
+        r = _finalize(cand, site_cfg, title_norm, page_norms, "byline_node", site_names)
+        if r:
+            return r
+
+    # 5) 可见文本 byline 正则（扫正文前部纯文本）
+    hay = signals.html[:120_000]
+    text_only = clean_text(re.sub(r"<[^>]+>", " ", hay))[:6000]
+    for m in _BYLINE_RE.finditer(text_only):
+        v = m.group(1)
+        if has_cjk(v) and len(v) > 8:
+            continue
+        r = _finalize(v, site_cfg, title_norm, page_norms, "byline_hint", site_names)
+        if r:
+            return r
+    for m in _BYLINE_EN_RE.finditer(text_only):
+        r = _finalize(m.group(1), site_cfg, title_norm, page_norms, "byline_en", site_names)
+        if r:
+            return r
+
+    # 6) 中文末尾记者署名（（记者 张三）/文｜张三）
+    nm = _cn_tail_sign(signals)
+    if nm:
+        r = _finalize(nm, site_cfg, title_norm, page_norms, "cn_tail_sign", site_names)
+        if r:
+            return r
+
+    # 7) 文章末尾通讯社署名（Elissa Mendes, The Canadian Press → 只取人名）
+    tail = _visible_tail(signals)
+    for nm in _wire_byline_candidates(tail):
+        r = _finalize(nm, site_cfg, title_norm, page_norms, "wire_byline", site_names)
+        if r:
+            return r
+
+    return "", "none", False