fix: remove keyword-based tag extraction, use AI auto-tag only

- extractTagsFromOrganization → no-op (keyword matching too imprecise)
- Tags now solely from Gemini auto-tag cron (industry/location/expertise/seniority)
- Migration 0051: backfill legacy tag data (category/normalized_value)

Author: iim0663418

workers/migrations/0051_backfill_keyword_tags.sql

@@ -0,0 +1,27 @@
+-- Migration 0051: Backfill auto_keyword tags missing category/normalized_value
+-- 修復 extractTagsFromOrganization 寫入的 tag 缺少 category/raw_value/normalized_value
+
+UPDATE card_tags
+SET category = 'keyword',
+    raw_value = tag,
+    normalized_value = tag
+WHERE tag_source = 'auto_keyword'
+  AND category IS NULL;
+
+-- 修復英文 key → 中文可讀標籤
+UPDATE card_tags SET tag = '政府機關', raw_value = '政府機關', normalized_value = '政府機關' WHERE tag = 'government' AND tag_source = 'auto_keyword';
+UPDATE card_tags SET tag = '企業法人', raw_value = '企業法人', normalized_value = '企業法人' WHERE tag = 'listed' AND tag_source = 'auto_keyword';
+UPDATE card_tags SET tag = '新創公司', raw_value = '新創公司', normalized_value = '新創公司' WHERE tag = 'startup' AND tag_source = 'auto_keyword';
+UPDATE card_tags SET tag = '非營利組織', raw_value = '非營利組織', normalized_value = '非營利組織' WHERE tag = 'ngo' AND tag_source = 'auto_keyword';
+
+-- tag_stats 也需要更新
+UPDATE tag_stats SET tag = '政府機關' WHERE tag = 'government';
+UPDATE tag_stats SET tag = '企業法人' WHERE tag = 'listed';
+UPDATE tag_stats SET tag = '新創公司' WHERE tag = 'startup';
+UPDATE tag_stats SET tag = '非營利組織' WHERE tag = 'ngo';
+
+-- 回滾方案（註解）
+-- UPDATE card_tags SET tag = 'government', raw_value = 'government', normalized_value = 'government' WHERE tag = '政府機關' AND tag_source = 'auto_keyword';
+-- UPDATE card_tags SET tag = 'listed', raw_value = 'listed', normalized_value = 'listed' WHERE tag = '企業法人' AND tag_source = 'auto_keyword';
+-- UPDATE card_tags SET tag = 'startup', raw_value = 'startup', normalized_value = 'startup' WHERE tag = '新創公司' AND tag_source = 'auto_keyword';
+-- UPDATE card_tags SET tag = 'ngo', raw_value = 'ngo', normalized_value = 'ngo' WHERE tag = '非營利組織' AND tag_source = 'auto_keyword';

workers/src/handlers/mcp/tools.ts

@@ -354,9 +354,9 @@ export async function toolSaveReceivedCard(
     if (tags.length > 0) {
       const tagStatements = tags.map(tag =>
         env.DB.prepare(`
-          INSERT OR IGNORE INTO card_tags (card_uuid, tag, tag_source, created_at)
-          VALUES (?, ?, 'auto_keyword', ?)
-        `).bind(cardUuid, tag, now)
+          INSERT OR IGNORE INTO card_tags (card_uuid, tag, category, raw_value, normalized_value, tag_source, created_at)
+          VALUES (?, ?, 'keyword', ?, ?, 'auto_keyword', ?)
+        `).bind(cardUuid, tag, tag, tag, now)
       );
       await env.DB.batch(tagStatements);
 

workers/src/handlers/user/received-cards/crud.ts

@@ -232,9 +232,9 @@ export async function handleSaveCard(request: Request, env: Env, ctx: ExecutionC
           for (const tag of tags) {
             statements.push(
               env.DB.prepare(`
-                INSERT OR IGNORE INTO card_tags (card_uuid, tag, tag_source, created_at)
-                VALUES (?, ?, 'auto_keyword', ?)
-              `).bind(cardUuid, tag, now)
+                INSERT OR IGNORE INTO card_tags (card_uuid, tag, category, raw_value, normalized_value, tag_source, created_at)
+                VALUES (?, ?, 'keyword', ?, ?, 'auto_keyword', ?)
+              `).bind(cardUuid, tag, tag, tag, now)
             );
           }
           await env.DB.batch(statements);

workers/src/utils/tags.ts

@@ -1,52 +1,14 @@
 /**
  * Tag Extraction Utilities
  *
- * BDD Spec: Tag Extraction Logic
- * Automatically extracts tags from business card data based on organization field
+ * Organization-based keyword tagging removed (2026-06-23):
+ * Keyword matching was too imprecise. AI auto-tagging (Gemini) via
+ * auto-tag-cards cron provides accurate industry/location/seniority/expertise.
  */
 
 /**
- * Company type keywords for tag extraction
- * Each type maps to an array of keywords to match
+ * @deprecated No-op. Kept for backward compatibility with callers.
  */
-const COMPANY_TYPE_KEYWORDS = {
-  government: ['政府', '部會', '機關', '局', '署', '處'],
-  listed: ['股份有限公司', '有限公司', 'Co., Ltd.', 'Inc.'],
-  startup: ['新創', '創業', 'Startup'],
-  ngo: ['基金會', '協會', '學會', '公會']
-};
-
-/**
- * Extract tags from organization field
- *
- * @param organization - The organization field from business card
- * @returns Array of matched tags
- *
- * @example
- * extractTagsFromOrganization("數位發展部") // ["government"]
- * extractTagsFromOrganization("台積電股份有限公司") // ["listed"]
- * extractTagsFromOrganization("AI 新創科技") // ["startup"]
- * extractTagsFromOrganization("台灣人工智慧協會") // ["ngo"]
- * extractTagsFromOrganization(null) // []
- */
-export function extractTagsFromOrganization(organization: string | null | undefined): string[] {
-  // Handle null, undefined, or empty string
-  if (!organization) {
-    return [];
-  }
-
-  const tags = new Set<string>();
-  const lowerOrg = organization.toLowerCase();
-
-  // Iterate through each company type and its keywords
-  for (const [type, keywords] of Object.entries(COMPANY_TYPE_KEYWORDS)) {
-    for (const keyword of keywords) {
-      if (lowerOrg.includes(keyword.toLowerCase())) {
-        tags.add(type);
-        break; // Each type should only be added once
-      }
-    }
-  }
-
-  return Array.from(tags);
+export function extractTagsFromOrganization(_organization: string | null | undefined): string[] {
+  return [];
 }