autosnippet
Version:
Extract code patterns into a knowledge base for AI coding assistants
185 lines (184 loc) • 7.19 kB
TypeScript
/**
* SearchEngine - 统一搜索引擎
*
* 三级搜索策略: keyword → FieldWeighted ranking → semantic(可选)
* 从 V1 SearchServiceV2 迁移，适配 V2 架构
*/
import Logger from '../../infrastructure/logging/Logger.js';
import { CoarseRanker } from './CoarseRanker.js';
import { MultiSignalRanker } from './MultiSignalRanker.js';
import type { DbRow, RankingContext, Scorer, SearchAiProvider, SearchCrossEncoder, SearchDb, SearchEngineOptions, SearchHybridRetriever, SearchOptions, SearchResponse, SearchResultItem, SearchVectorService, SearchVectorStore } from './SearchTypes.js';
export { BM25Scorer } from './BM25Scorer.js';
export { FieldWeightedScorer } from './FieldWeightedScorer.js';
export type { BM25DocMeta, DbRow, DocMeta, RankingContext, RrfHit, Scorer, ScorerResult, SearchAiProvider, SearchCrossEncoder, SearchDb, SearchEngineOptions, SearchHybridRetriever, SearchOptions, SearchResponse, SearchResultItem, SearchVectorService, SearchVectorStore, SlimSearchResult, VectorHit, } from './SearchTypes.js';
export { groupByKind, slimSearchResult } from './SearchTypes.js';
export { tokenize } from './tokenizer.js';
/**
* SearchEngine - 完整搜索服务
* 整合召回评分 + 关键词 + 可选 AI 增强
*/
export declare class SearchEngine {
#private;
_cache: Map<string, {
data: SearchResponse;
time: number;
}>;
_cacheMaxAge: number;
_coarseRanker: CoarseRanker;
_crossEncoder: SearchCrossEncoder | null;
_fusionRecallWeight: number;
_fusionSemanticWeight: number;
_indexed: boolean;
_lastIndexTime: string | null;
_multiSignalRanker: MultiSignalRanker;
_signalBus: import('../../infrastructure/signal/SignalBus.js').SignalBus | null;
aiProvider: SearchAiProvider | null;
db: SearchDb;
hybridRetriever: SearchHybridRetriever | null;
logger: ReturnType<typeof Logger.getInstance>;
scorer: Scorer;
vectorService: SearchVectorService | null;
vectorStore: SearchVectorStore | null;
constructor(db: SearchDb & {
getDb?: () => SearchDb;
}, options?: SearchEngineOptions);
/** 构建搜索索引 - 从数据库加载所有可搜索实体 */
buildIndex(): void;
/** 确保索引已构建（幂等），supply 给需要准确 stats 的调用方 */
ensureIndex(): void;
/**
* 统一搜索入口
* @param query 搜索关键词
* @param options {type, limit, mode, useAI}
*/
search(query: string, options?: SearchOptions): Promise<SearchResponse>;
/**
* 统一排序管线:
* 规范化 → [CrossEncoder 语义重排] → CoarseRanker (E-E-A-T 5维)
* → MultiSignalRanker (6信号) → 上下文加成
*
* CrossEncoder 仅在构造时传入 crossEncoderReranker 且 AI 可用时生效，
* 否则自动跳过（零额外开销）。
*/
_applyRanking(items: SearchResultItem[], query: string, context?: RankingContext): Promise<{
recallScore: number;
score: number;
id: string;
title?: string;
description?: string;
trigger?: string;
type?: string;
kind?: string;
status?: string;
language?: string;
category?: string;
content?: string;
code?: string;
headers?: string;
moduleName?: string;
knowledgeType?: string;
qualityScore?: number;
usageCount?: number;
authorityScore?: number;
tags?: string[] | string;
difficulty?: string;
updatedAt?: string | null;
createdAt?: string | null;
whenClause?: string;
doClause?: string;
rankerScore?: number;
coarseScore?: number;
contextScore?: number;
}[]>;
/**
* 将召回结果转换为 Ranker 所需格式（解析 content JSON、映射信号字段）
* 保留原始 content 供下游消费者使用
*/
_normalizeForRanking(items: SearchResultItem[]): SearchResultItem[];
/**
* 关键词搜索 - 直接 SQL LIKE
* 返回包含 kind 字段的完整结果，使用 ESCAPE 防止通配符注入
* 当 SQL LIKE 无结果时，降级到 FieldWeighted 搜索以提升自然语言查询的召回率
*/
_keywordSearch(query: string, type: string, limit: number): SearchResultItem[];
/**
* 加权字段搜索（FieldWeightedScorer）
* 增加 Title/Trigger 精确匹配 bonus — 当 query 出现在标题/触发词中时
* 给予额外分数加成，确保精确匹配的条目排名靠前
*/
_scorerSearch(query: string, type: string, limit: number): SearchResultItem[];
/**
* 语义搜索 - 需要 AI Provider 的 embed 功能
* 不可用时降级到 FieldWeighted 搜索
* @returns >}
*/
_semanticSearch(query: string, type: string, limit: number): Promise<{
items: SearchResultItem[];
actualMode: string;
}>;
/**
* 补充详细字段（content / description / trigger / delivery 字段）— 批量 IN 查询
* 用于向量搜索结果与 FieldWeighted 结果的一致性
*/
_supplementDetails(items: SearchResultItem[]): void;
/**
* 刷新索引（增量模式）
*
* 策略:
* 1. 如果尚未构建索引 → 全量 buildIndex()
* 2. 否则只加载 updatedAt > lastIndexTime 的条目 + 已删除(deprecated)条目
* - 新增/更新 → scorer.updateDocument()
* - 已删除 → scorer.removeDocument()
* 3. 清空缓存以确保搜索结果刷新
*
* @param [opts] - force=true 强制全量重建
*/
refreshIndex(opts?: {
force?: boolean;
}): void;
/**
* 从 DB 行构建索引文本
*
* 高价值字段（title, trigger）通过重复出现提升 TF 权重
* — title ×3, trigger ×2, description ×1.5（通过重复 token 实现）
* 这确保标题匹配的文档获得显著更高的分数
* 注：此逻辑主要服务于 BM25Scorer，FieldWeightedScorer 内部已有字段权重机制
*/
_buildDocText(r: DbRow): string;
/**
* 从 DB 行构建文档 meta
*/
_buildDocMeta(r: DbRow): {
type: string;
title: string | undefined;
trigger: string;
description: string;
contentText: string;
status: string | undefined;
knowledgeType: string | undefined;
kind: string;
language: string;
category: string;
updatedAt: string | null;
createdAt: string | null;
difficulty: string;
tags: string[];
usageCount: number;
authorityScore: number;
qualityScore: number;
};
/** 获取索引统计（如果尚未构建索引，自动触发构建） */
getStats(): {
indexed: boolean;
totalDocuments: number;
avgDocLength: number;
cacheSize: number;
uniqueTokens: number;
hasVectorStore: boolean;
hasVectorService: boolean;
hasAiProvider: boolean;
};
_getCache(key: string): SearchResponse | null;
_setCache(key: string, data: SearchResponse): void;
}
export default SearchEngine;