feat: add match visibility and semantic highlighting to hybrid search

- Add textMatches, textScore, semanticScore, matchSource to search results
- Add highlight() method for zero-config text + semantic highlighting
- Increase word indexing limit to 5000 (handles articles/chapters)
- Optimize findMatchingWords() with O(1) fast path for semantic-only results
- Add production safety limits (500 chunks for highlight)
- Add comprehensive tests for new features (35 tests)
- Update docs with match visibility and highlight() API
This commit is contained in:
David Snelling 2026-01-26 17:16:18 -08:00
parent b0ea2f3810
commit 4adba1b254
7 changed files with 1727 additions and 31 deletions

View file

@ -86,6 +86,25 @@ import { NounType, VerbType } from './types/graphTypes.js'
import { BrainyInterface } from './types/brainyInterface.js'
import type { IntegrationHub } from './integrations/core/IntegrationHub.js'
/**
* Stopwords for semantic highlighting (v7.8.0)
* These common words are skipped when highlighting individual words
* to focus on meaningful content words.
*/
const STOPWORDS = new Set([
'a', 'an', 'the', 'is', 'are', 'was', 'were', 'be', 'been',
'being', 'have', 'has', 'had', 'do', 'does', 'did', 'will', 'would', 'could', 'should',
'may', 'might', 'must', 'shall', 'can', 'to', 'of', 'in', 'for', 'on', 'with', 'at',
'by', 'from', 'as', 'into', 'through', 'during', 'before', 'after', 'above', 'below',
'between', 'under', 'again', 'further', 'then', 'once', 'here', 'there', 'when',
'where', 'why', 'how', 'all', 'each', 'few', 'more', 'most', 'other', 'some', 'such',
'no', 'nor', 'not', 'only', 'own', 'same', 'so', 'than', 'too', 'very', 'just',
'and', 'but', 'or', 'if', 'because', 'until', 'while', 'although', 'though',
'this', 'that', 'these', 'those', 'it', 'its', 'i', 'me', 'my', 'you', 'your',
'he', 'him', 'his', 'she', 'her', 'we', 'us', 'our', 'they', 'them', 'their',
'what', 'which', 'who', 'whom', 'whose', 'am'
])
/**
* The main Brainy class - Clean, Beautiful, Powerful
* REAL IMPLEMENTATION - No stubs, no mocks
@ -1921,25 +1940,49 @@ export class Brainy<T = any> implements BrainyInterface<T> {
return results
}
// Execute parallel searches for optimal performance
const searchPromises: Promise<Result<T>[]>[] = []
// Vector search component
if (params.query || params.vector) {
searchPromises.push(this.executeVectorSearch(params))
// v7.7.0: Zero-Config Hybrid Search
// Determine search mode: auto (default) combines text + semantic for query searches
const searchMode = params.searchMode || 'auto'
const limit = params.limit || 10
// Handle text-only query (user explicitly wants text search)
if (searchMode === 'text' && params.query && params.query.trim() !== '') {
results = await this.executeTextSearch(params.query, limit * 2)
}
// Proximity search component
if (params.near) {
searchPromises.push(this.executeProximitySearch(params))
// Handle semantic-only query (user explicitly wants vector search)
else if ((searchMode === 'semantic' || searchMode === 'vector') && (params.query || params.vector)) {
results = await this.executeVectorSearch(params)
}
// Execute searches in parallel
if (searchPromises.length > 0) {
const searchResults = await Promise.all(searchPromises)
for (const batch of searchResults) {
results.push(...batch)
}
// Handle explicit hybrid or auto mode with query
else if ((searchMode === 'auto' || searchMode === 'hybrid') && params.query && params.query.trim() !== '' && !params.vector) {
// Zero-config hybrid: combine text + semantic search with RRF fusion
const [textResults, semanticResults] = await Promise.all([
this.executeTextSearch(params.query, limit * 2),
this.executeVectorSearch(params)
])
// Use user-specified alpha or auto-detect based on query length
const alpha = params.hybridAlpha ?? this.autoAlpha(params.query)
// v7.8.0: Tokenize query for match visibility
const queryWords = this.metadataIndex.tokenize(params.query)
// RRF fusion combines both result sets with match visibility
results = await this.rrfFusion(textResults, semanticResults, alpha, queryWords)
}
// Handle direct vector search (no query text) - no hybrid needed
else if (params.vector && !params.query) {
results = await this.executeVectorSearch(params)
}
// Handle proximity search
else if (params.near) {
results = await this.executeProximitySearch(params)
}
// Execute parallel searches for additional criteria (proximity search in addition to query)
if (params.near && params.query) {
const proximityResults = await this.executeProximitySearch(params)
results.push(...proximityResults)
}
// Remove duplicate results from parallel searches
@ -2109,11 +2152,10 @@ export class Brainy<T = any> implements BrainyInterface<T> {
results.sort((a, b) => b.score - a.score)
}
const limit = params.limit || 10
const offset = params.offset || 0
const finalOffset = params.offset || 0
// Efficient pagination - only slice what we need
return results.slice(offset, offset + limit)
// Efficient pagination - only slice what we need (limit already defined above)
return results.slice(finalOffset, finalOffset + limit)
})
// Record performance for auto-tuning
@ -4662,6 +4704,160 @@ export class Brainy<T = any> implements BrainyInterface<T> {
return 1 - distance
}
/**
* Zero-config hybrid highlighting (v7.8.0)
*
* Returns both exact text matches AND semantically similar concepts.
* Perfect for UI highlighting at different levels:
* - matchType: 'text' = exact word match (highlight strongly)
* - matchType: 'semantic' = concept match (highlight softly)
*
* @param params.query - The search query
* @param params.text - The text to highlight (e.g., entity.data)
* @param params.granularity - 'word' | 'phrase' | 'sentence' (default: 'word')
* @param params.threshold - Minimum similarity for semantic matches (default: 0.5)
* @returns Array of highlights with text, score, position, and matchType
*
* @example
* ```typescript
* const highlights = await brain.highlight({
* query: "david the warrior",
* text: "David Smith is a brave fighter who battles dragons"
* })
* // Returns: [
* // { text: "David", score: 1.0, position: [0, 5], matchType: 'text' }, // Exact
* // { text: "fighter", score: 0.78, position: [25, 32], matchType: 'semantic' }, // Concept
* // { text: "battles", score: 0.72, position: [37, 44], matchType: 'semantic' } // Concept
* // ]
* ```
*/
async highlight(params: import('./types/brainy.types.js').HighlightParams): Promise<import('./types/brainy.types.js').Highlight[]> {
await this.ensureInitialized()
const { query, text, granularity = 'word', threshold = 0.5 } = params
if (!query || !text) {
return []
}
// Split text into chunks based on granularity
const allChunks = this.splitForHighlighting(text, granularity)
if (allChunks.length === 0) {
return []
}
// v7.8.0 Production safety: Limit chunks to prevent memory explosion
// At 500 words × 384 dimensions × 4 bytes = 768KB temp memory (acceptable)
// At 10,000 words = 15MB temp memory (too much for concurrent requests)
const MAX_HIGHLIGHT_CHUNKS = 500
const chunks = allChunks.slice(0, MAX_HIGHLIGHT_CHUNKS)
// Track all highlights (keyed by position to avoid duplicates)
const highlightMap = new Map<string, import('./types/brainy.types.js').Highlight>()
// === PHASE 1: Find exact text matches (score = 1.0, matchType = 'text') ===
const queryWords = this.metadataIndex.tokenize(query)
const queryWordsLower = new Set(queryWords.map(w => w.toLowerCase()))
for (const chunk of chunks) {
const chunkLower = chunk.text.toLowerCase().replace(/[^\w\s]/g, '')
if (queryWordsLower.has(chunkLower)) {
const key = `${chunk.position[0]}-${chunk.position[1]}`
highlightMap.set(key, {
text: chunk.text,
score: 1.0,
position: chunk.position,
matchType: 'text'
})
}
}
// === PHASE 2: Find semantic matches (score varies, matchType = 'semantic') ===
// Get query embedding
const queryVector = await this.embed(query)
// Batch embed all chunks (efficient!)
const chunkTexts = chunks.map(c => c.text)
const chunkVectors = await this.embedBatch(chunkTexts)
// Calculate semantic similarities
for (let i = 0; i < chunks.length; i++) {
const key = `${chunks[i].position[0]}-${chunks[i].position[1]}`
// Skip if already a text match (text matches take priority)
if (highlightMap.has(key)) continue
const distance = this.distance(queryVector, chunkVectors[i])
const similarity = 1 - distance
if (similarity >= threshold) {
highlightMap.set(key, {
text: chunks[i].text,
score: similarity,
position: chunks[i].position,
matchType: 'semantic'
})
}
}
// Sort by score descending (text matches will be first with score=1.0)
const highlights = Array.from(highlightMap.values())
return highlights.sort((a, b) => b.score - a.score)
}
/**
* Split text into chunks for highlighting (v7.8.0)
* @internal
*/
private splitForHighlighting(text: string, granularity: string): Array<{ text: string, position: [number, number] }> {
const results: Array<{ text: string, position: [number, number] }> = []
if (granularity === 'word') {
// Split on whitespace, track positions
const regex = /\S+/g
let match
while ((match = regex.exec(text)) !== null) {
// Skip stopwords
if (!STOPWORDS.has(match[0].toLowerCase())) {
results.push({ text: match[0], position: [match.index, match.index + match[0].length] })
}
}
} else if (granularity === 'sentence') {
// Split on sentence boundaries
const regex = /[^.!?]+[.!?]+/g
let match
while ((match = regex.exec(text)) !== null) {
results.push({ text: match[0].trim(), position: [match.index, match.index + match[0].length] })
}
// Handle text without sentence-ending punctuation
if (results.length === 0 && text.trim()) {
results.push({ text: text.trim(), position: [0, text.length] })
}
} else if (granularity === 'phrase') {
// Sliding window of 2-4 words
const words: Array<{ text: string, start: number, end: number }> = []
const regex = /\S+/g
let match
while ((match = regex.exec(text)) !== null) {
words.push({ text: match[0], start: match.index, end: match.index + match[0].length })
}
// Generate 2-4 word phrases
for (let windowSize = 2; windowSize <= 4; windowSize++) {
for (let i = 0; i <= words.length - windowSize; i++) {
const phraseWords = words.slice(i, i + windowSize)
const phraseText = phraseWords.map(w => w.text).join(' ')
const start = phraseWords[0].start
const end = phraseWords[phraseWords.length - 1].end
results.push({ text: phraseText, position: [start, end] })
}
}
}
return results
}
/**
* Get comprehensive index statistics
*
@ -5370,6 +5566,199 @@ export class Brainy<T = any> implements BrainyInterface<T> {
}
}
/**
* Execute text search using word index (v7.7.0)
*
* Performs keyword-based search using the __words__ index in MetadataIndexManager.
* Returns results ranked by word match count.
*
* @param query - Text query to search for
* @param limit - Maximum results to return
* @returns Array of Results with scores based on match count
*/
private async executeTextSearch(query: string, limit: number): Promise<Result<T>[]> {
const textMatches = await this.metadataIndex.getIdsForTextQuery(query)
if (textMatches.length === 0) return []
// Take top matches and load entities
const topMatches = textMatches.slice(0, limit * 2) // Get more for filtering
const ids = topMatches.map(m => m.id)
const entitiesMap = await this.batchGet(ids)
// Create results with scores based on match count
const maxMatches = topMatches[0]?.matchCount || 1
const results: Result<T>[] = []
for (const match of topMatches) {
const entity = entitiesMap.get(match.id)
if (entity) {
// Normalize score to 0-1 range based on match count
const score = match.matchCount / maxMatches
results.push(this.createResult(match.id, score, entity))
}
}
return results
}
/**
* Auto-detect optimal alpha for hybrid search (v7.7.0)
*
* Short queries (1-2 words) favor text search (lower alpha)
* Long queries (5+ words) favor semantic search (higher alpha)
*
* @param query - The search query
* @returns Alpha value between 0 (text only) and 1 (semantic only)
*/
private autoAlpha(query: string): number {
const wordCount = query.trim().split(/\s+/).filter(w => w.length > 0).length
if (wordCount <= 2) return 0.3 // Favor text for short queries
if (wordCount <= 5) return 0.5 // Balanced
return 0.7 // Favor semantic for long queries
}
/**
* Reciprocal Rank Fusion (RRF) for combining search results (v7.7.0)
*
* RRF is a proven fusion algorithm that:
* - Doesn't require score normalization
* - Handles different score distributions
* - Gives higher weight to top-ranked items
*
* Formula: score(d) = sum(1 / (k + rank(d))) for each list
*
* v7.8.0: Now includes match visibility (textMatches, textScore, semanticScore, matchSource)
*
* @param textResults - Results from text search
* @param semanticResults - Results from semantic search
* @param alpha - Weight for semantic (0=text only, 1=semantic only)
* @param queryWords - Original query words for match tracking
* @param k - RRF constant (default: 60, standard in literature)
* @returns Fused results sorted by combined score with match visibility
*/
private async rrfFusion(
textResults: Result<T>[],
semanticResults: Result<T>[],
alpha: number,
queryWords: string[],
k: number = 60
): Promise<Result<T>[]> {
// Track scores and match details per entity
interface MatchData {
rrf: number
textScore?: number
semanticScore?: number
textMatches: string[]
hasText: boolean
hasSemantic: boolean
}
const matchData = new Map<string, MatchData>()
const entityMap = new Map<string, Entity<T>>()
// Text contribution (1 - alpha weight)
const textWeight = 1 - alpha
textResults.forEach((r, rank) => {
const rrfScore = textWeight * (1 / (k + rank + 1))
const existing = matchData.get(r.id) || { rrf: 0, textMatches: [], hasText: false, hasSemantic: false }
existing.rrf += rrfScore
existing.textScore = r.score // Original text search score (0-1)
existing.hasText = true
matchData.set(r.id, existing)
if (r.entity) entityMap.set(r.id, r.entity)
})
// Semantic contribution (alpha weight)
semanticResults.forEach((r, rank) => {
const rrfScore = alpha * (1 / (k + rank + 1))
const existing = matchData.get(r.id) || { rrf: 0, textMatches: [], hasText: false, hasSemantic: false }
existing.rrf += rrfScore
existing.semanticScore = r.score // Original semantic search score (0-1)
existing.hasSemantic = true
matchData.set(r.id, existing)
if (r.entity) entityMap.set(r.id, r.entity)
})
// Sort by fused score
const sortedIds = Array.from(matchData.entries())
.sort((a, b) => b[1].rrf - a[1].rrf)
.map(([id, data]) => ({ id, data }))
// Build results - need to load any missing entities
const missingIds = sortedIds.filter(s => !entityMap.has(s.id)).map(s => s.id)
if (missingIds.length > 0) {
const loaded = await this.batchGet(missingIds)
for (const [id, entity] of loaded) {
entityMap.set(id, entity)
}
}
// v7.8.0 Performance: Build set of text result IDs for O(1) lookup
// This avoids re-extracting text for entities that weren't in text results
const textResultIds = new Set(textResults.map(r => r.id))
// Create final results with match visibility
const results: Result<T>[] = []
for (const { id, data } of sortedIds) {
const entity = entityMap.get(id)
if (entity) {
// Find which query words matched - uses fast path if entity wasn't in text results
const textMatches = this.findMatchingWords(entity, queryWords, textResultIds)
// Determine match source
let matchSource: 'text' | 'semantic' | 'both'
if (data.hasText && data.hasSemantic) {
matchSource = 'both'
} else if (data.hasText) {
matchSource = 'text'
} else {
matchSource = 'semantic'
}
// Create result with match visibility
const result = this.createResult(id, data.rrf, entity)
result.textMatches = textMatches
result.textScore = data.textScore
result.semanticScore = data.semanticScore
result.matchSource = matchSource
results.push(result)
}
}
return results
}
/**
* Find which query words match in an entity's text content (v7.8.0)
*
* Performance: O(query_words × text_length) - only called when needed
* At scale: Use textResultIds set for O(1) lookup instead of re-extracting
*
* @param entity - Entity to check
* @param queryWords - Words from the search query
* @param textResultIds - Optional: Set of IDs from text search (O(1) lookup)
* @returns Array of matching query words
*/
private findMatchingWords(
entity: Entity<T>,
queryWords: string[],
textResultIds?: Set<string>
): string[] {
// Fast path: if entity wasn't in text results, no words matched
if (textResultIds && !textResultIds.has(entity.id)) {
return []
}
// Slow path: extract text and check each word
// Only happens for entities that DID match text search
const textContent = this.metadataIndex.extractTextContent({
data: entity.data,
metadata: entity.metadata
}).toLowerCase()
return queryWords.filter(word => textContent.includes(word.toLowerCase()))
}
/**
* Apply graph constraints using O(1) GraphAdjacencyIndex - TRUE Triple Intelligence!
*/