mirror of
https://github.com/CyC2018/Interview-Notebook.git
synced 2026-11-19 13:03:47 +00:00
Update search.js
去掉“原理”“机制”这类泛泛的词再搜一次;多个词时只是文章名的词出现在小节标题中少加分;去掉编号后正好是搜索词的小节在前;RAG 与检索增强生成、集群与 cluster 互通。
This commit is contained in:
@@ -20,7 +20,7 @@
|
||||
// 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);
|
||||
prepared.push({ page: page, heading: heading, body: body, text: heading + ' ' + page + ' ' + body,
|
||||
lead: body.slice(0, 120), flatHeading: heading.replace(/ /g, ''), flatBody: body.replace(/ /g, ''), penalty: penalty, pageTitle: doc.title, path: doc.path, section: section });
|
||||
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 });
|
||||
});
|
||||
});
|
||||
preparedIndexes.set(index, prepared);
|
||||
@@ -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']
|
||||
['多版本并发控制', 'mvcc'], ['java 虚拟机', 'jvm'], ['检索增强生成', 'rag'], ['集群', 'cluster']
|
||||
];
|
||||
// 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 = [];
|
||||
@@ -101,7 +101,10 @@
|
||||
if (!term) return null;
|
||||
// 小节标题正好是搜索词时最相关,例如“死锁的必要条件”中的“必要条件”;标题用的是另一种说法时也算(搜 MVCC,标题是“多版本并发控制”)
|
||||
var head = firstHit(entry.heading, groups[i]) || term;
|
||||
total += (entry.heading === head ? 30 : 0) + (entry.heading.includes(head) ? (wholeWord(entry.heading, head) ? 40 : 25) : 0) +
|
||||
// 多个词时,就是文章名(或“ - ”分开的一段)的词只说明是哪篇文章,出现在小节标题中少加分(“spring aop”中的 spring)
|
||||
var headBonus = groups.length > 1 && entry.pageParts.indexOf(head) !== -1 ? 10 : wholeWord(entry.heading, head) ? 40 : 25;
|
||||
// 小节标题正好是搜索词时最相关;只有一个词时去掉编号(“三、”“2.”)再比
|
||||
total += (entry.heading === head || (groups.length === 1 && entry.bare === head) ? 30 : 0) + (entry.heading.includes(head) ? headBonus : 0) +
|
||||
(entry.page.includes(term) ? 15 : 0) + (entry.lead.includes(term) ? 10 : 0) + occurrences(entry.body, term);
|
||||
}
|
||||
var phrase = groups.length > 1 && normalizedQuery ? normalizedQuery.replace(/ /g, '') : '';
|
||||
@@ -116,6 +119,11 @@
|
||||
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;
|
||||
function stripGeneric(query) {
|
||||
var terms = splitTerms(query.replace(GENERIC, ' ')).filter(function (term) { return term.length > 1; });
|
||||
return terms.length ? terms.join(' ') : '';
|
||||
}
|
||||
// 仍然没有结果时按相邻两个字的词组匹配,至少命中七成;原文中连续出现的长片段(“四个条件”)额外加分
|
||||
function bigrams(terms) {
|
||||
var pairs = [];
|
||||
@@ -169,6 +177,9 @@
|
||||
// 去掉问法后的结果合并进来:“死锁的必要条件”原文只出现在学习路线中,正文的小节标题是“必要条件”
|
||||
var stripped = stripQuestion(normalizedQuery);
|
||||
if (stripped && stripped !== normalizedQuery) exact(stripped);
|
||||
// “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) {
|
||||
var terms = splitTerms(stripped || normalizedQuery);
|
||||
var pairs = bigrams(terms.map(function (term) { return alternatives(term)[0]; }));
|
||||
|
||||
Reference in New Issue
Block a user