feat(swift): E2 Intent Classification voltooid
Epic 2 - Intent Classification (10 SP): - E2.S1: Local classifier met regex patterns (<50ms) - E2.S2: Entity extraction (patientName, category, content) - E2.S3: AI fallback met Claude Haiku bij confidence <0.8 - E2.S4: POST /api/intent/classify API route Two-tier systeem: 1. Lokale regex classificatie (snel, offline) 2. AI fallback voor lage confidence of complexe input Bouwplan bijgewerkt naar v1.4 (31/68 SP, 46% done) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Colin Lit <colin@example.com>
This commit is contained in:
colinislit
233
lib/swift/entity-extractor.ts
Normal file
233
lib/swift/entity-extractor.ts
Normal file
@@ -0,0 +1,233 @@
|
||||
/**
|
||||
* Entity Extractor
|
||||
*
|
||||
* Extracts entities (patient name, category, content) from user input.
|
||||
*/
|
||||
|
||||
import type { VerpleegkundigCategory } from '@/lib/types/report';
|
||||
import type { ExtractedEntities, SwiftIntent } from './types';
|
||||
|
||||
// Category aliases mapping to canonical values
|
||||
const CATEGORY_ALIASES: Record<string, VerpleegkundigCategory> = {
|
||||
// Medicatie
|
||||
medicatie: 'medicatie',
|
||||
medicijn: 'medicatie',
|
||||
medicijnen: 'medicatie',
|
||||
med: 'medicatie',
|
||||
meds: 'medicatie',
|
||||
|
||||
// ADL
|
||||
adl: 'adl',
|
||||
verzorging: 'adl',
|
||||
zorg: 'adl',
|
||||
wassen: 'adl',
|
||||
eten: 'adl',
|
||||
douchen: 'adl',
|
||||
|
||||
// Gedrag
|
||||
gedrag: 'gedrag',
|
||||
gedrags: 'gedrag',
|
||||
stemming: 'gedrag',
|
||||
mood: 'gedrag',
|
||||
emotie: 'gedrag',
|
||||
|
||||
// Incident
|
||||
incident: 'incident',
|
||||
val: 'incident',
|
||||
gevallen: 'incident',
|
||||
ongeluk: 'incident',
|
||||
agressie: 'incident',
|
||||
|
||||
// Observatie
|
||||
observatie: 'observatie',
|
||||
obs: 'observatie',
|
||||
waarneming: 'observatie',
|
||||
opmerking: 'observatie',
|
||||
};
|
||||
|
||||
// Command words to strip from input
|
||||
const COMMAND_WORDS = [
|
||||
'notitie',
|
||||
'dagnotitie',
|
||||
'nieuwe',
|
||||
'schrijf',
|
||||
'rapporteer',
|
||||
'registreer',
|
||||
'zoek',
|
||||
'zoeken',
|
||||
'vind',
|
||||
'wie',
|
||||
'is',
|
||||
'waar',
|
||||
'info',
|
||||
'gegevens',
|
||||
'dossier',
|
||||
'overdracht',
|
||||
'dienst',
|
||||
'klaar',
|
||||
'afronden',
|
||||
'einde',
|
||||
'start',
|
||||
'begin',
|
||||
];
|
||||
|
||||
// Common Dutch first names for better name detection
|
||||
const COMMON_NAMES = new Set([
|
||||
'jan', 'piet', 'klaas', 'marie', 'anna', 'lisa', 'eva', 'emma', 'sophie',
|
||||
'thomas', 'lucas', 'daan', 'sem', 'liam', 'noah', 'julia', 'sara', 'lotte',
|
||||
'willem', 'johannes', 'cornelis', 'hendrik', 'maria', 'johanna', 'elisabeth',
|
||||
'peter', 'hans', 'henk', 'johan', 'bert', 'dick', 'kees', 'jaap', 'wim',
|
||||
'annie', 'bep', 'corrie', 'dinie', 'els', 'gerda', 'hanneke', 'ineke', 'joke',
|
||||
]);
|
||||
|
||||
/**
|
||||
* Extract entities from user input based on the detected intent.
|
||||
*/
|
||||
export function extractEntities(input: string, intent: SwiftIntent): ExtractedEntities {
|
||||
const trimmedInput = input.trim().toLowerCase();
|
||||
const entities: ExtractedEntities = {};
|
||||
|
||||
switch (intent) {
|
||||
case 'dagnotitie':
|
||||
return extractDagnotatieEntities(trimmedInput, input);
|
||||
case 'zoeken':
|
||||
return extractZoekenEntities(trimmedInput, input);
|
||||
case 'overdracht':
|
||||
// Overdracht doesn't need entity extraction
|
||||
return entities;
|
||||
default:
|
||||
return entities;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Extract entities for dagnotitie intent.
|
||||
* Patterns:
|
||||
* - "notitie jan medicatie" → name: jan, category: medicatie
|
||||
* - "jan medicatie gegeven" → name: jan, category: medicatie, content: gegeven
|
||||
* - "notitie medicatie jan" → name: jan, category: medicatie
|
||||
*/
|
||||
function extractDagnotatieEntities(lowerInput: string, originalInput: string): ExtractedEntities {
|
||||
const entities: ExtractedEntities = {};
|
||||
const words = lowerInput.split(/\s+/);
|
||||
|
||||
// Remove command words
|
||||
const filteredWords = words.filter(w => !COMMAND_WORDS.includes(w));
|
||||
|
||||
// Find category
|
||||
let categoryIndex = -1;
|
||||
for (let i = 0; i < filteredWords.length; i++) {
|
||||
const category = CATEGORY_ALIASES[filteredWords[i]];
|
||||
if (category) {
|
||||
entities.category = category;
|
||||
categoryIndex = i;
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
// Find name (word that's not a category and looks like a name)
|
||||
for (let i = 0; i < filteredWords.length; i++) {
|
||||
if (i === categoryIndex) continue;
|
||||
|
||||
const word = filteredWords[i];
|
||||
// Check if it's a known name or starts with uppercase in original
|
||||
if (isLikelyName(word, originalInput)) {
|
||||
entities.patientName = capitalizeFirst(word);
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
// Extract remaining content
|
||||
const contentWords = filteredWords.filter((w, i) => {
|
||||
if (i === categoryIndex) return false;
|
||||
if (entities.patientName && w === entities.patientName.toLowerCase()) return false;
|
||||
return true;
|
||||
});
|
||||
|
||||
if (contentWords.length > 0) {
|
||||
entities.content = contentWords.join(' ');
|
||||
}
|
||||
|
||||
return entities;
|
||||
}
|
||||
|
||||
/**
|
||||
* Extract entities for zoeken intent.
|
||||
* Patterns:
|
||||
* - "zoek jan" → name: jan
|
||||
* - "wie is marie" → name: marie
|
||||
* - "dossier piet" → name: piet
|
||||
*/
|
||||
function extractZoekenEntities(lowerInput: string, originalInput: string): ExtractedEntities {
|
||||
const entities: ExtractedEntities = {};
|
||||
const words = lowerInput.split(/\s+/);
|
||||
|
||||
// Remove command words
|
||||
const filteredWords = words.filter(w => !COMMAND_WORDS.includes(w));
|
||||
|
||||
// The remaining word(s) should be the name
|
||||
for (const word of filteredWords) {
|
||||
if (isLikelyName(word, originalInput)) {
|
||||
entities.patientName = capitalizeFirst(word);
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
// If no name found but there are remaining words, use the first one
|
||||
if (!entities.patientName && filteredWords.length > 0) {
|
||||
entities.patientName = capitalizeFirst(filteredWords[0]);
|
||||
}
|
||||
|
||||
return entities;
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if a word is likely a patient name.
|
||||
*/
|
||||
function isLikelyName(word: string, originalInput: string): boolean {
|
||||
// Check if it's a common name
|
||||
if (COMMON_NAMES.has(word.toLowerCase())) {
|
||||
return true;
|
||||
}
|
||||
|
||||
// Check if the word starts with uppercase in the original input
|
||||
const regex = new RegExp(`\\b${escapeRegex(word)}\\b`, 'i');
|
||||
const match = originalInput.match(regex);
|
||||
if (match && match[0][0] === match[0][0].toUpperCase()) {
|
||||
return true;
|
||||
}
|
||||
|
||||
// Single word that's not a category or command
|
||||
if (
|
||||
word.length >= 2 &&
|
||||
!CATEGORY_ALIASES[word] &&
|
||||
!COMMAND_WORDS.includes(word) &&
|
||||
/^[a-z]+$/i.test(word)
|
||||
) {
|
||||
return true;
|
||||
}
|
||||
|
||||
return false;
|
||||
}
|
||||
|
||||
/**
|
||||
* Capitalize the first letter of a string.
|
||||
*/
|
||||
function capitalizeFirst(str: string): string {
|
||||
return str.charAt(0).toUpperCase() + str.slice(1).toLowerCase();
|
||||
}
|
||||
|
||||
/**
|
||||
* Escape special regex characters.
|
||||
*/
|
||||
function escapeRegex(str: string): string {
|
||||
return str.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
|
||||
}
|
||||
|
||||
/**
|
||||
* Parse category from a string (with alias support).
|
||||
*/
|
||||
export function parseCategory(input: string): VerpleegkundigCategory | undefined {
|
||||
const lower = input.toLowerCase().trim();
|
||||
return CATEGORY_ALIASES[lower];
|
||||
}
|
||||
Reference in New Issue
Block a user