update

2026-07-19 19:18:13 +08:00
parent 847a80f470
commit 6ae3e3d1de
4 changed files with 426 additions and 43 deletions
--- a/src/engine/title_parser.rs
+++ b/src/engine/title_parser.rs
@@ -4,6 +4,12 @@
 use regex::Regex;
 use std::collections::HashSet;
 use lazy_static::lazy_static;
+use jieba_rs::Jieba;
+
+// 全局 jieba 分词器
+lazy_static! {
+    static ref JIEBA: Jieba = Jieba::new();
+}

 // ── 常量 ──

@@ -30,11 +36,48 @@ const STOP_WORDS: &[&str] = &[
 /// 保留词性（仅用于 jieba 分词场景，这里做占位符）
 const _KEEP_POS: &[&str] = &["n", "nr", "ns", "nz", "v", "vn", "a", "an", "eng", "x"];

-/// 工作室/平台排除名
+/// 工作室/平台排除名（大幅扩充）
 const EXCLUDED_NAMES: &[&str] = &[
+    // 工作室/平台
    "Pure Media", "XiuRen秀人网", "尤蜜", "魅妍社", "花漾",
    "瑞丝馆", "撸图", "情欲萝莉", "蜜桃社", "tgod",
    "MiStar", "BLUECAKE", "HuaYang", "YouMi", "秀人",
+    "XIUREN", "YouMi尤蜜", "IMiss爱蜜社", "MiiTao蜜桃社",
+    "MFStar模范学院", "TGOD推女神", "DKGirl御女郎",
+    "XINGYAN星颜社", "XIAOYU语画界", "YOUMI尤蜜荟",
+    "FAIRY CLUB", "FAIRY-CLUB", "FEILIN嗲囡囡",
+    "CANDY糖果画报", "YouWu尤物馆", "UXING优星馆",
+    "108TV", "108酱", "RuiSG瑞丝馆", "SIW斯文传媒",
+    "SLADY猎女神", "SSA丝社", "SJA佳爷", "TASHA系列",
+    "Ugirls爱尤物", "V Girls", "VGirl", "WingS影私荟",
+    "YALAYI雅拉伊", "ISHOW爱秀", "ISHOW",
+    "写真", "私房", "套图", "系列", "合集", "作品",
+    // 国家/地区
+    "日本", "韩国", "中国", "台湾", "香港", "泰国", "越南",
+    "新加坡", "马来西亚", "菲律宾", "印度尼西亚", "美国", "俄罗斯",
+    "东京", "首尔", "北京", "上海", "广州", "深圳", "成都", "重庆",
+    "曼谷", "胡志明", "新加坡市", "马尼拉", "雅加达",
+    // 非人名常见词
+    "模特", "美女", "女神", "小姐姐", "小仙女", "御姐", "萝莉",
+    "性感", "诱惑", "清纯", "甜美", "可爱", "迷人", "魅力",
+    "时尚", "潮流", "高清", "原版", "无水印", "完整版",
+    "比基尼", "泳装", "内衣", "丝袜", "制服", "旗袍", "睡衣",
+    "浴室", "浴缸", "床上", "沙发", "厨房", "阳台",
+    "室内", "户外", "街拍", "旅拍", "海边", "沙滩", "泳池",
+    "cosplay", "角色扮演", "动漫", "游戏",
+    "超短裙", "包臀裙", "紧身裙", "连衣裙", "牛仔裤", "热裤",
+    "高跟鞋", "运动鞋", "凉鞋",
+    "美腿", "美胸", "翘臀", "玉足", "美背", "香肩", "蛮腰",
+    "白皙", "肌肤", "纹身",
+    "日系", "韩系", "森系", "复古", "冷艳", "优雅", "知性",
+    "俏皮", "温柔", "妩媚", "野性", "热情", "慵懒", "迷离",
+    "唯美", "清新", "冷傲",
+    // 工作室前缀
+    "PureMedia", "BlueCake", "HuaYang", "XiuRen", "MiStar",
+    "YouMi", "MFStar", "TGOD", "DKGirl", "XINGYAN", "XIAOYU",
+    "YOUMI", "FAIRYCLUB", "FEILIN", "CANDY", "YouWu", "UXING",
+    "RuiSG", "SIW", "SLADY", "SSA", "SJA", "TASHA", "Ugirls",
+    "VGIRLS", "VGIRL", "WING", "YALAYI", "ISHOW",
 ];

 lazy_static! {
@@ -81,11 +124,13 @@ lazy_static! {
 // ── 人名提取 ──

 /// 从标题中提取模特/人物名称
-/// 核心策略：
-/// 1. 清理标题（移除[工作室]、编号、N photos等）
-/// 2. 尝试用分隔符(–—-)切分，第一部分常含人名
-/// 3. 扫描匹配"2-5中文+可选英文"模式的词
-/// 4. 结合上下文（后面是否跟"写真/私房/性感"等关键词）
+/// 核心策略（按优先级）：
+/// 1. cosplay 格式匹配（Name cosplay Character）
+/// 2. jieba 分词 + 词性标注（优先提取 nr 人名）
+/// 3. 清理标题（移除[工作室]、编号、N photos等）
+/// 4. 用分隔符(–—-)切分，第一部分常含人名
+/// 5. 扫描匹配"2-5中文+可选英文"模式的词
+/// 6. 结合上下文（后面是否跟"写真/私房/性感"等关键词）
 pub fn extract_person_name_from_title(title: &str) -> Option<String> {
    let title = title.trim();
    if title.is_empty() {
@@ -103,6 +148,12 @@ pub fn extract_person_name_from_title(title: &str) -> Option<String> {
        return Some(clean_name(caps[1].trim()));
    }

+    // 步骤2.5: jieba 分词 + 词性标注（最高优先级的中文人名提取）
+    // 只保留词性为 nr(人名)、n(名词)、nz(其它专名) 的词
+    if let Some(name) = extract_person_with_jieba(&t) {
+        return Some(name);
+    }
+
    // 步骤3: 移除 [工作室] 前缀
    t = RE_BRACKET_PREFIX.replace(&t, "").trim().to_string();

@@ -123,7 +174,7 @@ pub fn extract_person_name_from_title(title: &str) -> Option<String> {
    // 策略A: 检查第一部分（分隔符前的内容最可能包含人名）
    if let Some(first) = parts.first() {
        if let Some(name) = extract_name_from_segment(first) {
-            if !is_excluded_name(&name) {
+            if is_likely_person_name(&name) {
                return Some(name);
            }
        }
@@ -141,7 +192,7 @@ pub fn extract_person_name_from_title(title: &str) -> Option<String> {
        };

        if let Some(name) = extract_name_from_segment(part) {
-            if !is_excluded_name(&name) {
+            if is_likely_person_name(&name) {
                if has_context_after {
                    return Some(name); // 有关键词上下文，优先返回
                }
@@ -152,7 +203,7 @@ pub fn extract_person_name_from_title(title: &str) -> Option<String> {
    // 策略C: 直接在整个标题中搜索人名模式
    for cap in RE_CN_NAME.captures_iter(&t) {
        let name = clean_name(&cap[1]);
-        if !is_excluded_name(&name) {
+        if is_likely_person_name(&name) {
            return Some(name);
        }
    }
@@ -160,6 +211,39 @@ pub fn extract_person_name_from_title(title: &str) -> Option<String> {
    None
 }

+/// 用 jieba 分词提取人名（最高优先级策略）
+/// 只保留词性为 nr(人名)、n(名词)、nz(其它专名) 的词
+/// 同时结合二次验证过滤非人名
+fn extract_person_with_jieba(title: &str) -> Option<String> {
+    // 对标题做词性标注分词（true = 使用 HMM 新词发现）
+    let tags = JIEBA.tag(title, true);
+
+    // 按优先级收集候选词：nr > nz > n
+    let mut candidates: Vec<(String, usize)> = Vec::new(); // (词, 优先级)
+
+    for tag in &tags {
+        let word = tag.word.trim();
+        if word.len() < 2 { continue; }
+
+        let priority = match tag.tag.as_str() {
+            "nr" => 0,    // 人名（最高优先级）
+            "nrfg" => 1,  // 古体人名
+            "nrt" => 2,   // 音译人名
+            "nz" => 3,    // 其它专名
+            "n" => 4,     // 名词
+            _ => continue, // 其他词性跳过
+        };
+
+        if is_likely_person_name(word) {
+            candidates.push((word.to_string(), priority));
+        }
+    }
+
+    // 按优先级排序，取第一个
+    candidates.sort_by(|a, b| a.1.cmp(&b.1));
+    candidates.into_iter().next().map(|(name, _)| clean_name(&name))
+}
+
 /// 从一个文本片段中提取人名
 fn extract_name_from_segment(segment: &str) -> Option<String> {
    let seg = segment.trim();
@@ -213,6 +297,50 @@ pub fn is_excluded_name(name: &str) -> bool {
    EXCLUDED_LOWER.contains(&name.to_lowercase())
 }

+/// 二次验证：判断一个词是否"看起来像人名"
+/// 返回 true = 像人名，可以保留；false = 不像，应该排除
+pub fn is_likely_person_name(name: &str) -> bool {
+    let name = name.trim();
+    if name.is_empty() { return false; }
+    if name.len() < 2 { return false; }
+
+    // 先排除已知的非人名
+    if is_excluded_name(name) { return false; }
+
+    let name_lower = name.to_lowercase();
+
+    // 纯数字排除
+    if name.chars().all(|c| c.is_ascii_digit()) { return false; }
+
+    // 含数字太多的排除（超过一半是数字）
+    let digit_count = name.chars().filter(|c| c.is_ascii_digit()).count();
+    if digit_count * 2 > name.len() { return false; }
+
+    // 纯英文且长度 > 15 的排除（太长不像人名）
+    let has_cn = RE_CHINESE.is_match(name);
+    if !has_cn && name.len() > 15 { return false; }
+
+    // 以数字开头的排除（除非是纯数字的英文名）
+    if name.chars().next().map(|c| c.is_ascii_digit()).unwrap_or(false) { return false; }
+
+    // 含特殊字符过多的排除
+    let special_count = name.chars().filter(|c| {
+        !c.is_alphanumeric() && !RE_CHINESE.is_match(&c.to_string())
+    }).count();
+    if special_count > 2 { return false; }
+
+    // 已知后缀排除（如 "photos", "series", "collection" 等）
+    let known_suffix = ["photos", "series", "collection", "edition", "set", "vol", "album"];
+    for suffix in known_suffix {
+        if name_lower.ends_with(suffix) { return false; }
+    }
+
+    // 全部是大写字母且长度 > 6 的排除（通常是缩写/工作室名）
+    if name.chars().all(|c| c.is_ascii_uppercase()) && name.len() > 6 { return false; }
+
+    true
+}
+
 // ── 人名清洗 ──

 /// 清洗人名：移除前缀、括号内容、末尾标点
@@ -330,6 +458,139 @@ pub fn split_person_name_with_aliases(raw_name: &str) -> (String, Vec<(String, S
    (primary_name, aliases)
 }

+// ── 一键提取：人名 + 标签（单次 jieba 分词） ──
+
+/// 一次 jieba 分词，同时提取人名和标签
+/// 返回 (人名Option, 标签Vec)
+/// 优势：只做一次分词，省去重复开销
+pub fn extract_person_and_tags(title: &str) -> (Option<String>, Vec<(String, String)>) {
+    let title = title.trim();
+    if title.is_empty() {
+        return (None, Vec::new());
+    }
+
+    // 做一次词性标注分词
+    let tags = JIEBA.tag(title, true);
+
+    // ── 第一步：提取人名（优先级最高）──
+    let person_name = extract_person_from_tags(&tags);
+
+    // ── 第二步：提取名词标签（排除人名）──
+    let tag_list = extract_tags_from_jieba_result(&tags, person_name.as_deref());
+
+    (person_name, tag_list)
+}
+
+/// 从 jieba 分词结果中提取人名
+fn extract_person_from_tags(tags: &[jieba_rs::Token]) -> Option<String> {
+    let mut candidates: Vec<(String, usize)> = Vec::new();
+
+    for tag in tags {
+        let word = tag.word.trim();
+        if word.len() < 2 { continue; }
+
+        let priority = match tag.tag.as_str() {
+            "nr" => 0,    // 人名（最高优先级）
+            "nrfg" => 1,  // 古体人名
+            "nrt" => 2,   // 音译人名
+            "nz" => 3,    // 其它专名
+            "n" => 4,     // 名词
+            _ => continue,
+        };
+
+        if is_likely_person_name(word) {
+            candidates.push((word.to_string(), priority));
+        }
+    }
+
+    candidates.sort_by(|a, b| a.1.cmp(&b.1));
+    candidates.into_iter().next().map(|(name, _)| clean_name(&name))
+}
+
+/// 从 jieba 分词结果中提取标签（排除人名）
+fn extract_tags_from_jieba_result(
+    tags: &[jieba_rs::Token],
+    person_name: Option<&str>,
+) -> Vec<(String, String)> {
+    use std::collections::HashSet;
+
+    let mut result = Vec::new();
+    let mut seen = HashSet::new();
+
+    // 停用词
+    let stop_words: HashSet<&str> = STOP_WORDS.iter().copied().collect();
+
+    // 人名排除
+    let person_exclude: Vec<String> = person_name
+        .map(|n| {
+            let mut v = vec![n.to_lowercase()];
+            for part in n.split_whitespace() {
+                v.push(part.to_lowercase());
+            }
+            if let Some(cn) = extract_cn(n) {
+                v.push(cn.to_lowercase());
+            }
+            v
+        })
+        .unwrap_or_default();
+
+    for tag in tags {
+        let word = tag.word.trim();
+        if word.len() < 2 { continue; }
+
+        let word_lower = word.to_lowercase();
+
+        // 纯数字跳过
+        if word.chars().all(|c| c.is_ascii_digit()) { continue; }
+
+        // 停用词跳过
+        if stop_words.contains(word_lower.as_str()) { continue; }
+
+        // 人名跳过
+        if person_exclude.iter().any(|p| word_lower.contains(p)) { continue; }
+
+        // 去重
+        if !seen.insert(word_lower.clone()) { continue; }
+
+        // 按词性分类
+        let pos = tag.tag.as_str();
+        let category = match pos {
+            "n" | "ng" => "noun",
+            "nz" => "proper",
+            "ns" => "location",
+            "nt" => "organization",
+            "vn" => "noun_verb",
+            "a" | "ad" | "an" => "adj",
+            "v" | "vd" | "vg" => "verb",
+            "eng" => "english",
+            "x" => "other",
+            // 排除的词性
+            "nr" | "nrfg" | "nrt" => continue, // 人名
+            "m" | "mq" => continue,
+            "q" | "qv" | "qt" => continue,
+            "r" | "rr" | "rz" => continue,
+            "c" => continue,
+            "p" => continue,
+            "u" | "ul" | "uv" | "ud" | "ug" | "uj" => continue,
+            "d" => continue,
+            "e" | "y" | "o" => continue,
+            "w" => continue,
+            "t" | "tg" => continue,
+            "s" => continue,
+            "f" => continue,
+            "b" | "bl" => continue,
+            "z" => continue,
+            _ => continue,
+        };
+
+        if category == "english" && word.len() < 3 { continue; }
+
+        result.push((word.to_string(), category.to_string()));
+    }
+
+    result
+}
+
 // ── 标签提取 ──

 /// 从标题中提取标签（排除人名部分）