Add L0 index and anchor UI updates

modules/story-summary/vector/llm/query-expansion.js

@@ -0,0 +1,102 @@
+// ═══════════════════════════════════════════════════════════════════════════
+// query-expansion.js - 完整输入，不截断
+// ═══════════════════════════════════════════════════════════════════════════
+
+import { callLLM, parseJson } from './llm-service.js';
+import { xbLog } from '../../../../core/debug-core.js';
+import { filterText } from '../utils/text-filter.js';
+
+const MODULE_ID = 'query-expansion';
+const SESSION_ID = 'xb6';
+
+const SYSTEM_PROMPT = `你是检索词生成器。根据最近对话，输出用于检索历史剧情的关键词。
+
+只输出JSON：
+{"e":["显式人物/地名"],"i":["隐含人物/情绪/话题"],"q":["检索短句"]}
+
+规则：
+- e: 对话中明确提到的人名/地名，1-4个
+- i: 推断出的相关人物/情绪/话题，1-5个
+- q: 用于向量检索的短句，2-3个，每个15字内
+- 关注：正在讨论什么、涉及谁、情绪氛围`;
+
+/**
+ * Query Expansion
+ * @param {Array} messages - 完整消息数组（最后2-3轮）
+ */
+export async function expandQuery(messages, options = {}) {
+    const { timeout = 6000 } = options;
+
+    if (!messages?.length) {
+        return { entities: [], implicit: [], queries: [] };
+    }
+
+    // 完整格式化，不截断
+    const input = messages.map(m => {
+        const speaker = m.is_user ? '用户' : (m.name || '角色');
+        const text = filterText(m.mes || '').trim();
+        return `【${speaker}】\n${text}`;
+    }).join('\n\n');
+
+    const T0 = performance.now();
+
+    try {
+        const response = await callLLM([
+            { role: 'system', content: SYSTEM_PROMPT },
+            { role: 'user', content: input },
+        ], {
+            temperature: 0.15,
+            max_tokens: 250,
+            timeout,
+            sessionId: SESSION_ID,
+        });
+
+        const parsed = parseJson(response);
+        if (!parsed) {
+            xbLog.warn(MODULE_ID, 'JSON解析失败', response?.slice(0, 200));
+            return { entities: [], implicit: [], queries: [] };
+        }
+
+        const result = {
+            entities: Array.isArray(parsed.e) ? parsed.e.slice(0, 5) : [],
+            implicit: Array.isArray(parsed.i) ? parsed.i.slice(0, 6) : [],
+            queries: Array.isArray(parsed.q) ? parsed.q.slice(0, 4) : [],
+        };
+
+        xbLog.info(MODULE_ID, `完成 (${Math.round(performance.now() - T0)}ms) e=${result.entities.length} i=${result.implicit.length} q=${result.queries.length}`);
+        return result;
+
+    } catch (e) {
+        xbLog.error(MODULE_ID, '调用失败', e);
+        return { entities: [], implicit: [], queries: [] };
+    }
+}
+
+// 缓存
+const cache = new Map();
+const CACHE_TTL = 300000;
+
+function hashMessages(messages) {
+    const text = messages.slice(-2).map(m => (m.mes || '').slice(0, 100)).join('|');
+    let h = 0;
+    for (let i = 0; i < text.length; i++) h = ((h << 5) - h + text.charCodeAt(i)) | 0;
+    return h.toString(36);
+}
+
+export async function expandQueryCached(messages, options = {}) {
+    const key = hashMessages(messages);
+    const cached = cache.get(key);
+    if (cached && Date.now() - cached.time < CACHE_TTL) return cached.result;
+
+    const result = await expandQuery(messages, options);
+    if (result.entities.length || result.queries.length) {
+        if (cache.size > 50) cache.delete(cache.keys().next().value);
+        cache.set(key, { result, time: Date.now() });
+    }
+    return result;
+}
+
+export function buildSearchText(expansion) {
+    return [...(expansion.entities || []), ...(expansion.implicit || []), ...(expansion.queries || [])]
+        .filter(Boolean).join(' ');
+}