diff --git a/website/builder/test_search.cjs b/website/builder/test_search.cjs index 3c3fb1c8..b0d47847 100644 --- a/website/builder/test_search.cjs +++ b/website/builder/test_search.cjs @@ -68,6 +68,21 @@ assert.equal(search(generic, 'synchronized 原理')[0].id, 'sync', '“原理” assert.equal(search(generic, 'spring aop')[0].id, 'aop', '文章名中的词出现在小节标题里少加分'); assert.equal(search(generic, 'rag 是什么')[0].id, 'rag', 'RAG 与检索增强生成互通,去掉编号后正好是搜索词的小节在前'); assert.equal(search(generic, '进程和线程的区别')[0].id, 'both', '多个词时,只匹配其中一个词的小节不算正好匹配'); +const golden = { documents: [ + { title: '设计模式 - 单例', path: '/notes/单例', sections: [{ title: '单例(Singleton)', id: 'single', text: '确保一个类只有一个实例' }] }, + { title: '攻击技术', path: '/notes/攻击', sections: [{ title: '一、跨站脚本攻击', id: 'xss', text: '攻击者把脚本注入页面' }] }, + { title: '数据库', path: '/notes/库', sections: [ + { title: '四、隔离级别', id: 'level', text: '未提交读、提交读、可重复读、可串行化' }, + { title: '参考资料', id: 'refs', text: '事务隔离级别 的文档' }, + ] }, + { title: '容器 - 进阶', path: '/notes/容器 - 进阶', sections: [{ title: 'ConcurrentHashMap', id: 'deep', text: 'ConcurrentHashMap 的源码' }] }, + { title: '容器', path: '/notes/容器', sections: [{ title: 'ConcurrentHashMap', id: 'main', text: 'ConcurrentHashMap 的源码' }] }, +] }; +assert.equal(search(golden, '单例模式')[0].id, 'single', '“单例模式”也按“单例”搜'); +assert.equal(search(golden, 'XSS')[0].id, 'xss', 'XSS 与跨站脚本互通'); +assert.equal(search(golden, '事务隔离级别')[0].id, 'level', '完全匹配的只有参考资料时,相近的小节排在前面'); +assert(!search(golden, '事务隔离级别').fuzzy, '有完全匹配的结果时不提示“没有完全匹配”'); +assert.equal(search(golden, 'concurrenthashmap')[0].id, 'main', '主笔记排在进阶篇之前'); const sample = { documents: [{ title: '数据库', path: '/notes/示例', sections: [ { title: '事务', id: '事务', text: '一致性 '.repeat(20) }, { title: '一致性', id: '一致性', text: 'SQL 与数据库' }, diff --git a/website/templates/search.js b/website/templates/search.js index e93dd3e3..b351e65f 100644 --- a/website/templates/search.js +++ b/website/templates/search.js @@ -18,7 +18,7 @@ var heading = normalize(section.title); var body = normalize(section.text); // 学习路线只是阅读导读,附录是实验过程,排在讲解正文之后 - var penalty = (/^学习路线/.test(doc.title) ? 45 : 0) + (/^a\d+ /.test(heading) ? 40 : 0); + var penalty = (/^学习路线/.test(doc.title) ? 45 : 0) + (/^a\d+ /.test(heading) ? 40 : 0) + (/ - 进阶$/.test(doc.title) ? 8 : 0) + (heading === '参考资料' ? 30 : 0); prepared.push({ page: page, heading: heading, body: body, text: heading + ' ' + page + ' ' + body, pageParts: page.split(' - ').map(function (x) { return x.trim(); }), bare: heading.replace(/^(?:[一二三四五六七八九十]+、|[a-z]?\d+(?:\.\d+)*\.?\s)\s*/, ''), lead: body.slice(0, 120), flatHeading: heading.replace(/ /g, ''), flatBody: body.replace(/ /g, ''), penalty: penalty, pageTitle: doc.title, path: doc.path, section: section }); }); @@ -58,7 +58,7 @@ var SYNONYMS = [ ['垃圾收集', '垃圾回收'], ['i/o', 'io'], ['哈希', '散列', 'hash'], ['新生代', '年轻代'], ['重排序', '指令重排'], ['首部', '头部'], ['幻读', '幻影读'], ['脏读', '读脏数据'], ['b+tree', 'b+树', 'b+ 树'], ['多路复用', 'i/o 复用', '复用'], - ['多版本并发控制', 'mvcc'], ['java 虚拟机', 'jvm'], ['检索增强生成', 'rag'], ['集群', 'cluster'] + ['多版本并发控制', 'mvcc'], ['java 虚拟机', 'jvm'], ['检索增强生成', 'rag'], ['集群', 'cluster'], ['跨站脚本', 'xss'], ['跨站请求伪造', 'csrf'], ['运行时数据区域', '内存区域', '内存结构'] ]; // 中英文混写的术语(“b+ 树”“i/o 复用”)在切词前换成一个整体,否则会被切成“b+”和“树”两个词分别匹配 var MIXED = []; @@ -78,6 +78,8 @@ // 词中含有另一种说法时也替换,例如“散列冲突”同时按“哈希冲突”搜索 function alternatives(term) { var result = [term]; + // “单例模式”“观察者模式”:正文常只写“单例”“观察者” + if (/^[㐀-鿿]{2,}模式$/.test(term) && term !== '设计模式') result.push(term.slice(0, -2)); SYNONYMS.forEach(function (group) { group.forEach(function (word) { if (term.indexOf(word) === -1) return; @@ -119,7 +121,7 @@ var terms = splitTerms(normalize(query.replace(QUESTION, ' '))).filter(function (term) { return term.length > 1; }); return terms.length ? terms.join(' ') : ''; } - var GENERIC = /(底层)?(原理|机制|实现方式|实现|作用|流程|过程|优缺点|应用场景|使用场景|场景|特点|详解|介绍|用法)/g; + var GENERIC = /(底层)?(原理|机制|实现方式|实现|作用|流程|过程|优缺点|应用场景|使用场景|场景|特点|详解|介绍|用法|理论)/g; function stripGeneric(query) { var terms = splitTerms(query.replace(GENERIC, ' ')).filter(function (term) { return term.length > 1; }); return terms.length ? terms.join(' ') : ''; @@ -136,14 +138,14 @@ }); return pairs; } - function fuzzyScore(entry, pairs, segments) { + function fuzzyScore(entry, pairs, segments, ratio) { var hit = 0, total = 0; pairs.forEach(function (pair) { if (!entry.text.includes(pair)) return; hit++; total += (entry.heading.includes(pair) ? 12 : 0) + (entry.page.includes(pair) ? 4 : 0) + Math.min(occurrences(entry.body, pair), 3); }); - if (hit < Math.max(2, Math.ceil(pairs.length * .7))) return null; + if (hit < Math.max(2, Math.ceil(pairs.length * (ratio || .7)))) return null; segments.forEach(function (segment) { if (segment.length > 2 && entry.text.includes(segment)) total += segment.length * 6; }); @@ -180,16 +182,19 @@ // “原理”“机制”这类泛泛的词正文中不一定出现(“synchronized 原理”),去掉后再搜一次;结果比带着这些词命中的少一个词的分 var core = stripGeneric(stripped || normalizedQuery); if (core && core !== normalizedQuery && core !== stripped) exact(core); - if (!results.length) { + if (results.length < 3) { + var exactCount = results.length; var terms = splitTerms(stripped || normalizedQuery); var pairs = bigrams(terms.map(function (term) { return alternatives(term)[0]; })); var ascii = terms.filter(function (term) { return !CJK.test(term); }).map(alternatives); if (pairs.length >= 2) { - fuzzy = true; - used = pairs.concat([].concat.apply([], ascii)); + // 已有完全匹配的结果时,相近的只是补在后面,不提示“没有完全匹配” + fuzzy = !exactCount; + used = (exactCount ? used : []).concat(pairs, [].concat.apply([], ascii)); collect(function (entry) { if (ascii.some(function (group) { return !firstHit(entry.text, group); })) return null; - return fuzzyScore(entry, pairs, terms); + // 补在完全匹配之后时放宽到六成:“事务隔离级别”中的“务隔”在正文中不一定连着出现 + return fuzzyScore(entry, pairs, terms, exactCount ? .6 : .7); }); } // 只剩英文词时(“HashMap和Hashtable的区别”中的“区别”找不到同一节),只按英文词搜