Update core/qualitative_engine.py

core/qualitative_engine.py

@@ -1,6 +1,8 @@
 # -*- coding: utf-8 -*-
 import json
 import logging
+import re
+import streamlit as st
 
 class QualitativeEngine:
     def __init__(self, agent):
@@ -9,24 +11,36 @@ class QualitativeEngine:
     def resolve_qualitative_data(self, topic, keywords, raw_interviews=""):
         """
         核心逻辑：解析真实访谈数据，或在无数据时模拟 ASQ 级别的定性数据结构表。
+        引入流式状态机，防止 WebSocket 假死超时。
         """
         if raw_interviews and len(raw_interviews) > 100:
             logging.info("📝 检测到用户上传访谈记录，正在通过 DeepSeek-Reasoner 进行深度主题萃取...")
             prompt = f"Topic: {topic}. Extract the 15 most powerful, representative quotes from the following interview/field text. Organize into a Gioia-style coding structure. Return ONLY a valid JSON array of objects with keys: 'theme' (2nd-order), 'concept' (1st-order), and 'quote' (exact words)."
-            response = self.agent.call_deepseek("You are an expert qualitative analyst.", f"DATA:\n{raw_interviews[:30000]}", model_name="deepseek-reasoner")
+            
+            # [修复点 1] 改用流式渲染，保活网关连接
+            with st.status("🧠 正在使用 R1 模型深度萃取访谈语录 (耗时较长，请观察下方思考流以防断连)...", expanded=True) as status:
+                ph = st.empty()
+                response = ph.write_stream(self.agent.stream_deepseek("You are an expert qualitative analyst.", f"DATA:\n{raw_interviews[:30000]}", model_name="deepseek-reasoner"))
+                status.update(label="✅ 访谈主题萃取完成！", state="complete", expanded=False)
+                
+            # 清理流式输出中带有的 think 标签
+            response = re.sub(r'<think>.*?</think>', '', response, flags=re.DOTALL).strip()
         else:
             logging.info("⚙️ 未检测到真实访谈记录，正在通过蒙特卡洛引擎模拟高质量定性编码树防幻觉...")
             prompt = f"Topic: {topic}. Keywords: {keywords}. As an ASQ researcher, simulate a rigorous set of qualitative data for this study. Provide 12 highly realistic, human-sounding representative interview quotes. Group them into emergent 1st-order concepts and 2nd-order themes. Return ONLY a valid JSON array of objects with keys: 'theme', 'concept', 'quote'."
-            response = self.agent.call_deepseek("You are an expert qualitative analyst.", prompt, model_name="deepseek-chat")
+            
+            # [修复点 2] 模拟过程同样流式化
+            with st.status("⚙️ 正在通过 V3 模型极速模拟定性编码树...", expanded=True) as status:
+                ph = st.empty()
+                response = ph.write_stream(self.agent.stream_deepseek("You are an expert qualitative analyst.", prompt, model_name="deepseek-chat"))
+                status.update(label="✅ 模拟定性编码树构建完成！", state="complete", expanded=False)
         
         data = self.agent._extract_json_from_text(response)
         
-        # 容错处理
         if not data:
             logging.error("❌ 定性数据 JSON 解析失败，回退至默认模拟数据。")
-            return "### Table: Qualitative Coding Structure\n| 2nd-order Theme | 1st-order Concept | Representative Quote |\n| :--- | :--- | :--- |\n| Sensemaking | Initial Confusion | \"We really didn't know how to adapt to the new framework.\" [QUOTE-01] |"
+            return "### Table: Qualitative Coding Structure\n| 2nd-order Theme | 1st-order Concept | Representative Quote |\n| :--- | :--- | :--- |\n| Sensemaking | Initial Confusion | \"We really didn't know how to adapt to the new framework.\" [QUOTE-01] |", "⚠️ 数据解析出现偏差，已启用降级保护预案。"
         
-        # 兼容 JSON 格式 (字典或列表)
         if isinstance(data, dict):
             items = data.get("data", data.get("quotes", []))
         elif isinstance(data, list):
@@ -34,7 +48,6 @@ class QualitativeEngine:
         else:
             items = []
 
-        # 转换为 Markdown 表格，供流水线作为事实锚点注入
         md = "### Table: Qualitative Coding Structure (Data Evidence)\n| 2nd-order Theme | 1st-order Concept | Representative Quote |\n| :--- | :--- | :--- |\n"
         for i, item in enumerate(items[:15]):
             q = str(item.get("quote", "")).replace('"', "'").replace("\n", " ")