f2bc2d54cb
Implements Rule 46: Ontology-Driven Cache Segmentation Semantic Cache Enhancements: - Add institutionSubtype, recordSetType, wikidataEntity to ExtractedEntities - Add extractionMethod field to track vocabulary vs regex extraction - Implement async extractEntitiesWithVocabulary() using term log - Maintain sync regex fallback for cache key generation (<5ms) Build Pipeline: - Add prebuild hook to regenerate types-vocab.json from LinkML schemas - Extract vocabulary from *Type.yaml and *Types.yaml schema files - Generate GLAMORCUBESFIXPHDNT code mappings automatically New Script: - scripts/extract-types-vocab.ts - Extracts vocabulary from LinkML schemas - Supports --skip-embeddings flag for faster builds - Outputs to apps/archief-assistent/public/types-vocab.json This enables richer cache segmentation using ontology-derived subtypes (e.g., 'MUNICIPAL_ARCHIVE', 'ART_MUSEUM') instead of just top-level GLAMORCUBESFIXPHDNT codes.
494 lines
18 KiB
JavaScript
494 lines
18 KiB
JavaScript
#!/usr/bin/env node
|
|
/**
|
|
* extract-types-vocab.ts
|
|
*
|
|
* Extracts vocabulary from LinkML *Type.yaml and *Types.yaml schema files
|
|
* and generates embeddings for two-tier semantic routing.
|
|
*
|
|
* Output: apps/archief-assistent/public/types-vocab.json
|
|
*
|
|
* Usage:
|
|
* npx tsx scripts/extract-types-vocab.ts
|
|
* npx tsx scripts/extract-types-vocab.ts --skip-embeddings # Skip embedding generation
|
|
*
|
|
* See: .opencode/rules/ontology-driven-cache-segmentation.md
|
|
*/
|
|
|
|
import { readFileSync, writeFileSync, readdirSync, existsSync, mkdirSync } from 'node:fs';
|
|
import { join, dirname } from 'node:path';
|
|
import { fileURLToPath } from 'node:url';
|
|
import { parse as parseYaml } from 'yaml';
|
|
|
|
// ESM compatibility for __dirname
|
|
const __filename = fileURLToPath(import.meta.url);
|
|
const __dirname = dirname(__filename);
|
|
|
|
// ============================================================================
|
|
// Configuration
|
|
// ============================================================================
|
|
|
|
const SCHEMA_DIR = join(__dirname, '../schemas/20251121/linkml/modules/classes');
|
|
const OUTPUT_FILE = join(__dirname, '../apps/archief-assistent/public/types-vocab.json');
|
|
const EMBEDDING_API_URL = process.env.EMBEDDING_API_URL || 'http://localhost:8000/api/embed';
|
|
|
|
// GLAMORCUBESFIXPHDNT code mapping
|
|
const TYPE_FILE_TO_CODE: Record<string, string> = {
|
|
'ArchiveOrganizationType': 'A',
|
|
'BioCustodianType': 'B',
|
|
'CommercialOrganizationType': 'C',
|
|
'DigitalPlatformType': 'D',
|
|
'EducationProviderType': 'E',
|
|
'FeatureCustodianType': 'F',
|
|
'GalleryType': 'G',
|
|
'HolySacredSiteType': 'H',
|
|
'IntangibleHeritageGroupType': 'I',
|
|
'LibraryType': 'L',
|
|
'MuseumType': 'M',
|
|
'NonProfitType': 'N',
|
|
'OfficialInstitutionType': 'O',
|
|
'PersonalCollectionType': 'P',
|
|
'ResearchOrganizationType': 'R',
|
|
'HeritageSocietyType': 'S',
|
|
'TasteScentHeritageType': 'T',
|
|
'UnspecifiedType': 'U',
|
|
'MixedCustodianType': 'X',
|
|
};
|
|
|
|
// ============================================================================
|
|
// Types
|
|
// ============================================================================
|
|
|
|
interface TermLogEntry {
|
|
typeCode: string;
|
|
typeName: string;
|
|
subtypeName?: string;
|
|
recordSetType?: string;
|
|
wikidata?: string;
|
|
lang: string;
|
|
}
|
|
|
|
interface SubtypeInfo {
|
|
className: string;
|
|
wikidata?: string;
|
|
accumulatedTerms: string;
|
|
keywords: Record<string, string[]>;
|
|
}
|
|
|
|
interface TypeInfo {
|
|
code: string;
|
|
className: string;
|
|
baseWikidata?: string;
|
|
accumulatedTerms: string;
|
|
keywords: Record<string, string[]>;
|
|
subtypes: Record<string, SubtypeInfo>;
|
|
}
|
|
|
|
interface RecordSetTypeInfo {
|
|
className: string;
|
|
accumulatedTerms: string;
|
|
keywords: Record<string, string[]>;
|
|
}
|
|
|
|
interface TypesVocabulary {
|
|
version: string;
|
|
schemaVersion: string;
|
|
embeddingModel: string;
|
|
embeddingDimensions: number;
|
|
tier1Embeddings: Record<string, number[]>;
|
|
tier2Embeddings: Record<string, Record<string, number[]>>;
|
|
termLog: Record<string, TermLogEntry>;
|
|
institutionTypes: Record<string, TypeInfo>;
|
|
recordSetTypes: Record<string, RecordSetTypeInfo>;
|
|
}
|
|
|
|
interface ParsedClass {
|
|
className: string;
|
|
description?: string;
|
|
keywords?: string[];
|
|
structuredAliases?: Array<{ literal_form: string; in_language?: string }>;
|
|
wikidataEntity?: string;
|
|
isSubtypeOf?: string;
|
|
}
|
|
|
|
// ============================================================================
|
|
// YAML Parsing
|
|
// ============================================================================
|
|
|
|
function parseYamlFile(filePath: string): Record<string, unknown> | null {
|
|
try {
|
|
const content = readFileSync(filePath, 'utf-8');
|
|
return parseYaml(content);
|
|
} catch (error) {
|
|
console.warn(`Warning: Could not parse ${filePath}: ${error}`);
|
|
return null;
|
|
}
|
|
}
|
|
|
|
function extractClassesFromYaml(yamlData: Record<string, unknown>): ParsedClass[] {
|
|
const classes: ParsedClass[] = [];
|
|
const classesSection = yamlData.classes as Record<string, unknown> | undefined;
|
|
|
|
if (!classesSection) return classes;
|
|
|
|
for (const [className, classDef] of Object.entries(classesSection)) {
|
|
if (typeof classDef !== 'object' || classDef === null) continue;
|
|
|
|
const classData = classDef as Record<string, unknown>;
|
|
|
|
// Skip abstract base classes (except the main Type class)
|
|
if (classData.abstract === true && !className.endsWith('Type')) continue;
|
|
|
|
const parsed: ParsedClass = {
|
|
className,
|
|
description: classData.description as string | undefined,
|
|
keywords: classData.keywords as string[] | undefined,
|
|
structuredAliases: classData.structured_aliases as Array<{ literal_form: string; in_language?: string }> | undefined,
|
|
isSubtypeOf: classData.is_a as string | undefined,
|
|
};
|
|
|
|
// Extract wikidata entity from slot_usage or mappings
|
|
const slotUsage = classData.slot_usage as Record<string, unknown> | undefined;
|
|
if (slotUsage?.wikidata_entity) {
|
|
const wdSlot = slotUsage.wikidata_entity as Record<string, unknown>;
|
|
parsed.wikidataEntity = wdSlot.equals_string as string | undefined;
|
|
}
|
|
|
|
// Check exact_mappings for Wikidata
|
|
const exactMappings = classData.exact_mappings as string[] | undefined;
|
|
if (exactMappings) {
|
|
const wdMapping = exactMappings.find(m => m.startsWith('wd:') || m.startsWith('wikidata:'));
|
|
if (wdMapping) {
|
|
parsed.wikidataEntity = wdMapping.replace(/^(wd:|wikidata:)/, '');
|
|
}
|
|
}
|
|
|
|
// Check broad_mappings for Wikidata
|
|
const broadMappings = classData.broad_mappings as string[] | undefined;
|
|
if (broadMappings && !parsed.wikidataEntity) {
|
|
const wdMapping = broadMappings.find(m => m.startsWith('wd:'));
|
|
if (wdMapping) {
|
|
parsed.wikidataEntity = wdMapping.replace('wd:', '');
|
|
}
|
|
}
|
|
|
|
classes.push(parsed);
|
|
}
|
|
|
|
return classes;
|
|
}
|
|
|
|
function extractKeywordsFromClass(parsedClass: ParsedClass): Record<string, string[]> {
|
|
const keywords: Record<string, string[]> = {};
|
|
|
|
// 1. Extract from keywords array (usually language-agnostic, assume Dutch/English)
|
|
if (parsedClass.keywords) {
|
|
keywords['nl'] = keywords['nl'] || [];
|
|
keywords['en'] = keywords['en'] || [];
|
|
for (const kw of parsedClass.keywords) {
|
|
// Simple heuristic: Dutch words often have Dutch-specific patterns
|
|
const isDutch = /[ij]|sch|cht|aa|ee|oo|uu/i.test(kw);
|
|
if (isDutch) {
|
|
keywords['nl'].push(kw.toLowerCase());
|
|
} else {
|
|
keywords['en'].push(kw.toLowerCase());
|
|
}
|
|
}
|
|
}
|
|
|
|
// 2. Extract from structured_aliases (language-tagged)
|
|
if (parsedClass.structuredAliases) {
|
|
for (const alias of parsedClass.structuredAliases) {
|
|
const lang = alias.in_language || 'en';
|
|
keywords[lang] = keywords[lang] || [];
|
|
keywords[lang].push(alias.literal_form.toLowerCase());
|
|
}
|
|
}
|
|
|
|
// 3. Convert class name to keywords
|
|
// MunicipalArchive -> ["municipal archive", "municipal", "archive"]
|
|
const classNameWords = parsedClass.className
|
|
.replace(/([A-Z])/g, ' $1')
|
|
.trim()
|
|
.toLowerCase()
|
|
.split(/\s+/);
|
|
|
|
keywords['en'] = keywords['en'] || [];
|
|
keywords['en'].push(classNameWords.join(' '));
|
|
|
|
return keywords;
|
|
}
|
|
|
|
function accumulateTerms(keywords: Record<string, string[]>): string {
|
|
const allTerms: string[] = [];
|
|
for (const terms of Object.values(keywords)) {
|
|
allTerms.push(...terms);
|
|
}
|
|
return [...new Set(allTerms)].join(' ');
|
|
}
|
|
|
|
// ============================================================================
|
|
// Embedding Generation
|
|
// ============================================================================
|
|
|
|
async function generateEmbedding(text: string, skipEmbeddings: boolean): Promise<number[]> {
|
|
if (skipEmbeddings) {
|
|
// Return empty placeholder
|
|
return [];
|
|
}
|
|
|
|
try {
|
|
const response = await fetch(EMBEDDING_API_URL, {
|
|
method: 'POST',
|
|
headers: { 'Content-Type': 'application/json' },
|
|
body: JSON.stringify({ text }),
|
|
});
|
|
|
|
if (!response.ok) {
|
|
console.warn(`Embedding API error: ${response.status}`);
|
|
return [];
|
|
}
|
|
|
|
const data = await response.json();
|
|
return data.embedding || [];
|
|
} catch (error) {
|
|
console.warn(`Embedding generation failed: ${error}`);
|
|
return [];
|
|
}
|
|
}
|
|
|
|
// ============================================================================
|
|
// Main Processing
|
|
// ============================================================================
|
|
|
|
async function processTypeFiles(): Promise<TypesVocabulary> {
|
|
const skipEmbeddings = process.argv.includes('--skip-embeddings');
|
|
|
|
console.log('🔍 Scanning schema directory:', SCHEMA_DIR);
|
|
console.log(`📊 Embedding generation: ${skipEmbeddings ? 'SKIPPED' : 'ENABLED'}`);
|
|
|
|
const vocabulary: TypesVocabulary = {
|
|
version: new Date().toISOString(),
|
|
schemaVersion: '20251121',
|
|
embeddingModel: 'paraphrase-multilingual-MiniLM-L12-v2',
|
|
embeddingDimensions: 384,
|
|
tier1Embeddings: {},
|
|
tier2Embeddings: {},
|
|
termLog: {},
|
|
institutionTypes: {},
|
|
recordSetTypes: {},
|
|
};
|
|
|
|
// Find all *Type.yaml files (base types)
|
|
const files = readdirSync(SCHEMA_DIR);
|
|
const typeFiles = files.filter(f => f.endsWith('Type.yaml') && !f.endsWith('Types.yaml'));
|
|
const typesFiles = files.filter(f => f.endsWith('Types.yaml'));
|
|
|
|
console.log(`\n📁 Found ${typeFiles.length} Type files and ${typesFiles.length} Types files`);
|
|
|
|
// Process base Type files
|
|
for (const file of typeFiles) {
|
|
const typeName = file.replace('.yaml', '');
|
|
const code = TYPE_FILE_TO_CODE[typeName];
|
|
|
|
if (!code) {
|
|
console.log(` ⏭️ Skipping ${typeName} (not in GLAMORCUBESFIXPHDNT)`);
|
|
continue;
|
|
}
|
|
|
|
console.log(`\n📄 Processing ${typeName} (${code})`);
|
|
|
|
const filePath = join(SCHEMA_DIR, file);
|
|
const yamlData = parseYamlFile(filePath);
|
|
if (!yamlData) continue;
|
|
|
|
const classes = extractClassesFromYaml(yamlData);
|
|
const baseClass = classes.find(c => c.className === typeName);
|
|
|
|
if (!baseClass) {
|
|
console.log(` ⚠️ No base class found in ${file}`);
|
|
continue;
|
|
}
|
|
|
|
// Initialize type info
|
|
const typeInfo: TypeInfo = {
|
|
code,
|
|
className: typeName,
|
|
baseWikidata: baseClass.wikidataEntity,
|
|
accumulatedTerms: '',
|
|
keywords: extractKeywordsFromClass(baseClass),
|
|
subtypes: {},
|
|
};
|
|
|
|
// Look for corresponding Types file (subtypes)
|
|
const subtypesFilePath = join(SCHEMA_DIR, file.replace('Type.yaml', 'Types.yaml'));
|
|
|
|
if (existsSync(subtypesFilePath)) {
|
|
console.log(` 📂 Processing subtypes from ${subtypesFilePath.split('/').pop()}`);
|
|
const subtypesYaml = parseYamlFile(subtypesFilePath);
|
|
if (subtypesYaml) {
|
|
const subtypeClasses = extractClassesFromYaml(subtypesYaml);
|
|
|
|
for (const subclass of subtypeClasses) {
|
|
// Convert CamelCase to UPPER_SNAKE_CASE
|
|
const subtypeName = subclass.className
|
|
.replace(/([a-z])([A-Z])/g, '$1_$2')
|
|
.toUpperCase();
|
|
const subtypeKeywords = extractKeywordsFromClass(subclass);
|
|
|
|
const subtypeInfo: SubtypeInfo = {
|
|
className: subclass.className,
|
|
wikidata: subclass.wikidataEntity,
|
|
accumulatedTerms: accumulateTerms(subtypeKeywords),
|
|
keywords: subtypeKeywords,
|
|
};
|
|
|
|
typeInfo.subtypes[subtypeName] = subtypeInfo;
|
|
|
|
// Add to term log
|
|
for (const [lang, terms] of Object.entries(subtypeKeywords)) {
|
|
for (const term of terms) {
|
|
vocabulary.termLog[term] = {
|
|
typeCode: code,
|
|
typeName,
|
|
subtypeName,
|
|
wikidata: subclass.wikidataEntity,
|
|
lang,
|
|
};
|
|
}
|
|
}
|
|
|
|
console.log(` ✓ ${subclass.className}: ${Object.values(subtypeKeywords).flat().length} terms`);
|
|
}
|
|
}
|
|
}
|
|
|
|
// Accumulate all terms for this type (base + all subtypes)
|
|
const allTypeTerms: string[] = [];
|
|
allTypeTerms.push(accumulateTerms(typeInfo.keywords));
|
|
for (const subtype of Object.values(typeInfo.subtypes)) {
|
|
allTypeTerms.push(subtype.accumulatedTerms);
|
|
}
|
|
typeInfo.accumulatedTerms = [...new Set(allTypeTerms.join(' ').split(' '))].join(' ');
|
|
|
|
// Add base type keywords to term log
|
|
for (const [lang, terms] of Object.entries(typeInfo.keywords)) {
|
|
for (const term of terms) {
|
|
vocabulary.termLog[term] = {
|
|
typeCode: code,
|
|
typeName,
|
|
lang,
|
|
};
|
|
}
|
|
}
|
|
|
|
vocabulary.institutionTypes[code] = typeInfo;
|
|
console.log(` ✅ ${typeName}: ${Object.keys(typeInfo.subtypes).length} subtypes, ${typeInfo.accumulatedTerms.split(' ').length} total terms`);
|
|
}
|
|
|
|
// Process RecordSetTypes files
|
|
console.log('\n📁 Processing RecordSetTypes files...');
|
|
const recordSetTypesFiles = files.filter(f => f.endsWith('RecordSetTypes.yaml'));
|
|
|
|
for (const file of recordSetTypesFiles) {
|
|
const filePath = join(SCHEMA_DIR, file);
|
|
const yamlData = parseYamlFile(filePath);
|
|
if (!yamlData) continue;
|
|
|
|
const classes = extractClassesFromYaml(yamlData);
|
|
|
|
for (const cls of classes) {
|
|
// Skip abstract base classes
|
|
if (cls.className.endsWith('RecordSetType') && !cls.className.includes('Fonds') &&
|
|
!cls.className.includes('Series') && !cls.className.includes('Collection')) {
|
|
continue;
|
|
}
|
|
|
|
// Convert CamelCase to UPPER_SNAKE_CASE
|
|
const rstName = cls.className
|
|
.replace(/([a-z])([A-Z])/g, '$1_$2')
|
|
.toUpperCase();
|
|
const keywords = extractKeywordsFromClass(cls);
|
|
|
|
const rstInfo: RecordSetTypeInfo = {
|
|
className: cls.className,
|
|
accumulatedTerms: accumulateTerms(keywords),
|
|
keywords,
|
|
};
|
|
|
|
vocabulary.recordSetTypes[rstName] = rstInfo;
|
|
|
|
// Add to term log
|
|
for (const [lang, terms] of Object.entries(keywords)) {
|
|
for (const term of terms) {
|
|
vocabulary.termLog[term] = {
|
|
typeCode: 'A', // Most record set types are archive-related
|
|
typeName: 'ArchiveOrganizationType',
|
|
recordSetType: rstName,
|
|
lang,
|
|
};
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
console.log(` ✅ Extracted ${Object.keys(vocabulary.recordSetTypes).length} record set types`);
|
|
|
|
// Generate Tier 1 embeddings (Types file level)
|
|
console.log('\n🧮 Generating Tier 1 embeddings (Types files)...');
|
|
for (const [code, typeInfo] of Object.entries(vocabulary.institutionTypes)) {
|
|
const embedding = await generateEmbedding(typeInfo.accumulatedTerms, skipEmbeddings);
|
|
vocabulary.tier1Embeddings[typeInfo.className] = embedding;
|
|
console.log(` ✓ ${typeInfo.className}: ${embedding.length} dimensions`);
|
|
}
|
|
|
|
// Generate Tier 2 embeddings (individual subtypes)
|
|
console.log('\n🧮 Generating Tier 2 embeddings (subtypes)...');
|
|
for (const [code, typeInfo] of Object.entries(vocabulary.institutionTypes)) {
|
|
vocabulary.tier2Embeddings[code] = {};
|
|
|
|
for (const [subtypeName, subtypeInfo] of Object.entries(typeInfo.subtypes)) {
|
|
const embedding = await generateEmbedding(subtypeInfo.accumulatedTerms, skipEmbeddings);
|
|
vocabulary.tier2Embeddings[code][subtypeName] = embedding;
|
|
}
|
|
|
|
console.log(` ✓ ${typeInfo.className}: ${Object.keys(typeInfo.subtypes).length} subtype embeddings`);
|
|
}
|
|
|
|
return vocabulary;
|
|
}
|
|
|
|
// ============================================================================
|
|
// Main Entry Point
|
|
// ============================================================================
|
|
|
|
async function main() {
|
|
console.log('═══════════════════════════════════════════════════════════════');
|
|
console.log(' TypesVocabulary Extraction Script');
|
|
console.log(' Ontology-Driven Cache Segmentation (Rule 46)');
|
|
console.log('═══════════════════════════════════════════════════════════════\n');
|
|
|
|
const vocabulary = await processTypeFiles();
|
|
|
|
// Ensure output directory exists
|
|
const outputDir = dirname(OUTPUT_FILE);
|
|
if (!existsSync(outputDir)) {
|
|
mkdirSync(outputDir, { recursive: true });
|
|
}
|
|
|
|
// Write output
|
|
writeFileSync(OUTPUT_FILE, JSON.stringify(vocabulary, null, 2));
|
|
|
|
console.log('\n═══════════════════════════════════════════════════════════════');
|
|
console.log(' Summary');
|
|
console.log('═══════════════════════════════════════════════════════════════');
|
|
console.log(` 📊 Institution Types: ${Object.keys(vocabulary.institutionTypes).length}`);
|
|
console.log(` 📊 Total Subtypes: ${Object.values(vocabulary.institutionTypes).reduce((sum, t) => sum + Object.keys(t.subtypes).length, 0)}`);
|
|
console.log(` 📊 Record Set Types: ${Object.keys(vocabulary.recordSetTypes).length}`);
|
|
console.log(` 📊 Term Log Entries: ${Object.keys(vocabulary.termLog).length}`);
|
|
console.log(` 📊 Tier 1 Embeddings: ${Object.keys(vocabulary.tier1Embeddings).length}`);
|
|
console.log(` 📊 Tier 2 Embeddings: ${Object.values(vocabulary.tier2Embeddings).reduce((sum, t) => sum + Object.keys(t).length, 0)}`);
|
|
console.log(`\n ✅ Output written to: ${OUTPUT_FILE}`);
|
|
console.log('═══════════════════════════════════════════════════════════════\n');
|
|
}
|
|
|
|
main().catch(console.error);
|