Add address detection with explicit context (Option A)

Conservative approach - only detects addresses with explicit labels:
- "Address:", "Shipping Address:", "Billing Address:"
- "Ship to:", "Deliver to:", "Send to:"
- "Location:", "Residence:"
- P.O. Box patterns
- Arabic label (عنوان)

Added to both standalone detector (24 types) and web app.
Added 2 new tests for address detection.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: iYassr

docs/app/index.html

@@ -1270,6 +1270,44 @@ <h3>Sanitized</h3>
         }
       }
 
+      // 24. Addresses (with explicit context only - conservative approach)
+      const addressPatterns = [
+        // Explicit address labels
+        /\b(?:address|mailing\s+address|shipping\s+address|billing\s+address|home\s+address|work\s+address|street\s+address|physical\s+address|postal\s+address)[:\s]+([^\n\r]{10,100})/gi,
+        // Ship to / Deliver to patterns
+        /\b(?:ship\s+to|deliver\s+to|send\s+to|mail\s+to)[:\s]+([^\n\r]{10,100})/gi,
+        // Location patterns
+        /\b(?:location|residence)[:\s]+([^\n\r]{10,100})/gi,
+        // P.O. Box patterns
+        /\b(P\.?O\.?\s*Box\s+\d+(?:[,\s]+[A-Za-z\s]+)?(?:[,\s]+\d{5})?)/gi,
+        // Arabic address label
+        /\b(?:عنوان|العنوان)[:\s]+([^\n\r]{10,100})/g
+      ];
+      for (const pattern of addressPatterns) {
+        pattern.lastIndex = 0;
+        while ((match = pattern.exec(text)) !== null) {
+          let addressText = (match[1] || match[0]).trim();
+          // Clean up
+          addressText = addressText
+            .replace(/[,;.]+$/, '')
+            .replace(/\s+(phone|tel|fax|email|contact|mobile).*$/i, '')
+            .trim();
+          // Skip if too short or no number and less than 3 words
+          if (addressText.length < 10) continue;
+          if (!/\d/.test(addressText) && addressText.split(/\s+/).length < 3) continue;
+
+          const addressStart = match[0].indexOf(addressText);
+          const start = match.index + (addressStart >= 0 ? addressStart : 0);
+          addEntity({
+            text: addressText,
+            type: 'address',
+            start: start,
+            end: start + addressText.length,
+            confidence: 90
+          });
+        }
+      }
+
       return entities.sort((a, b) => a.start - b.start);
     }
 

electron/services/detector.ts

@@ -45,7 +45,7 @@ export interface NEREntity {
   /** The detected text content */
   text: string
   /** Entity type classification */
-  type: 'person' | 'financial' | 'credit_card' | 'iban' | 'phone' | 'email' | 'ip' | 'url' | 'domain' | 'saudi_id' | 'ssn' | 'passport' | 'dob' | 'mac_address' | 'api_key' | 'license_plate' | 'medical_record' | 'drivers_license' | 'gps' | 'vin' | 'company_code'
+  type: 'person' | 'financial' | 'credit_card' | 'iban' | 'phone' | 'email' | 'ip' | 'url' | 'domain' | 'saudi_id' | 'ssn' | 'passport' | 'dob' | 'mac_address' | 'api_key' | 'license_plate' | 'medical_record' | 'drivers_license' | 'gps' | 'vin' | 'company_code' | 'address'
   /** Start position in source text (0-indexed) */
   start: number
   /** End position in source text (exclusive) */
@@ -252,6 +252,11 @@ export async function extractEntities(text: string, userCustomNames?: string[]):
   detectCompanyCodes(text, addEntity)
   logDebug('Company codes detection complete', { found: entities.length - before })
 
+  // 23. Extract addresses (with explicit context only)
+  before = countBefore()
+  detectAddresses(text, addEntity)
+  logDebug('Addresses detection complete', { found: entities.length - before })
+
   // Deduplicate entities (same position)
   const seen = new Set<string>()
   const uniqueEntities = entities.filter((e) => {
@@ -1583,6 +1588,70 @@ function detectCompanyCodes(
   }
 }
 
+/**
+ * Detects addresses with explicit context keywords (conservative approach).
+ * Only detects addresses that are explicitly labeled to minimize false positives.
+ *
+ * Supported contexts:
+ * - "Address:", "Shipping address:", "Billing address:"
+ * - "Ship to:", "Deliver to:", "Send to:"
+ * - "Location:", "Residence:", "Home:"
+ * - "P.O. Box", "PO Box"
+ *
+ * @internal
+ */
+function detectAddresses(
+  text: string,
+  addEntity: (entity: NEREntity) => void
+): void {
+  // Context patterns that indicate an address follows
+  const addressContextPatterns = [
+    // Explicit address labels - capture until end of line or next label
+    /\b(?:address|mailing\s+address|shipping\s+address|billing\s+address|home\s+address|work\s+address|street\s+address|physical\s+address|postal\s+address|residential\s+address)[:\s]+([^\n\r]{10,100})/gi,
+    // Ship to / Deliver to patterns
+    /\b(?:ship\s+to|deliver\s+to|send\s+to|mail\s+to)[:\s]+([^\n\r]{10,100})/gi,
+    // Location patterns
+    /\b(?:location|residence|domicile)[:\s]+([^\n\r]{10,100})/gi,
+    // P.O. Box patterns (very common in Saudi/GCC)
+    /\b(P\.?O\.?\s*Box\s+\d+(?:[,\s]+[A-Za-z\s]+)?(?:[,\s]+\d{5})?)/gi,
+    // Saudi/GCC format with building/street
+    /\b(?:عنوان|العنوان)[:\s]+([^\n\r]{10,100})/g,
+  ]
+
+  for (const pattern of addressContextPatterns) {
+    pattern.lastIndex = 0
+    let match: RegExpExecArray | null
+    while ((match = pattern.exec(text)) !== null) {
+      // Get the captured address part or full match for P.O. Box
+      let addressText = (match[1] || match[0]).trim()
+
+      // Clean up trailing punctuation and common end markers
+      addressText = addressText
+        .replace(/[,;.]+$/, '')
+        .replace(/\s+(phone|tel|fax|email|contact|mobile).*$/i, '')
+        .trim()
+
+      // Skip if too short (likely not a real address)
+      if (addressText.length < 10) continue
+
+      // Skip if it looks like just a name or single word
+      if (!/\d/.test(addressText) && addressText.split(/\s+/).length < 3) continue
+
+      // Find actual position of address text in the match
+      const addressStart = match[0].indexOf(addressText)
+      const start = match.index + (addressStart >= 0 ? addressStart : 0)
+
+      addEntity({
+        text: addressText,
+        type: 'address',
+        start: start,
+        end: start + addressText.length,
+        confidence: 90
+      })
+    }
+  }
+}
+
 /**
  * Detects person names from text (returns only custom names).
  *

tests/web-app.spec.ts

@@ -287,4 +287,27 @@ test.describe('Web App - Browser Version', () => {
     await expect(table).toContainText('iban')
     await expect(table).toContainText('financial')
   })
+
+  test('detects addresses with explicit context', async ({ page }) => {
+    await page.click('#toggleTextInput')
+    await page.fill('#textInput', 'Address: 123 Main Street, Boston, MA 02101')
+    await page.click('#analyzeTextBtn')
+
+    await page.waitForSelector('#step2:not(.hidden)', { timeout: 10000 })
+
+    const table = page.locator('#detectionTable')
+    await expect(table).toContainText('address')
+    await expect(table).toContainText('123 Main Street')
+  })
+
+  test('detects P.O. Box addresses', async ({ page }) => {
+    await page.click('#toggleTextInput')
+    await page.fill('#textInput', 'Send mail to P.O. Box 12345, Riyadh')
+    await page.click('#analyzeTextBtn')
+
+    await page.waitForSelector('#step2:not(.hidden)', { timeout: 10000 })
+
+    const table = page.locator('#detectionTable')
+    await expect(table).toContainText('P.O. Box 12345')
+  })
 })