Refine search query. (#5235)

### What problem does this PR solve?

#5173
#5214

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)

api/db/services/dialog_service.py

@@ -876,12 +876,16 @@ def reasoning(chunk_info: dict, question: str, chat_mdl: LLMBundle, embd_mdl: LL
         "Once you have all the information you need, continue your reasoning.\n\n"
         "-- Example --\n"
         "Question: \"Find the minimum number of vertices in a Steiner tree that includes all specified vertices in a given tree.\"\n"
-        "Assistant thinking steps:\n"
-        "- I need to understand what a Steiner tree is and how to compute the minimum number of vertices required to include all specified vertices in a given tree.\n\n"
         "Assistant:\n"
-        f"{BEGIN_SEARCH_QUERY}Minimum Steiner Tree problem in trees{END_SEARCH_QUERY}\n\n"
+        "  - I need to understand what a Steiner tree is.\n\n" 
-        "(System returns processed information from relevant web pages)\n\n"
+        f"    {BEGIN_SEARCH_QUERY}What's Steiner tree{END_SEARCH_QUERY}\n\n"
-        "Assistant continues reasoning with the new information...\n\n"
+        f"    {BEGIN_SEARCH_RESULT}\n(System returns processed information from relevant web pages)\n{END_SEARCH_RESULT}\n\n"
+        "User:\nContinues reasoning with the new information.\n\n"
+        "Assistant:\n"
+        "  - I need to understand what the difference between minimum number of vertices and edges in the Steiner tree is.\n\n" 
+        f"    {BEGIN_SEARCH_QUERY}What's the difference between minimum number of vertices and edges in the Steiner tree{END_SEARCH_QUERY}\n\n"
+        f"    {BEGIN_SEARCH_RESULT}\n(System returns processed information from relevant web pages)\n{END_SEARCH_RESULT}\n\n"
+        "User:\nContinues reasoning with the new information...\n\n"
         "**Remember**:\n"
         f"- You have a dataset to search, so you just provide a proper search query.\n"
         f"- Use {BEGIN_SEARCH_QUERY} to request a dataset search and end with {END_SEARCH_QUERY}.\n"
@@ -943,7 +947,7 @@ def reasoning(chunk_info: dict, question: str, chat_mdl: LLMBundle, embd_mdl: LL
 
         query_think = ""
         if msg_hisotry[-1]["role"] != "user":
-            msg_hisotry.append({"role": "user", "content": "Continues reasoning with the new information...\n"})
+            msg_hisotry.append({"role": "user", "content": "Continues reasoning with the new information.\n"})
         for ans in chat_mdl.chat_streamly(reason_prompt, msg_hisotry, {"temperature": 0.7}):
             ans = re.sub(r"<think>.*</think>", "", ans, flags=re.DOTALL)
             if not ans:
@@ -954,15 +958,13 @@ def reasoning(chunk_info: dict, question: str, chat_mdl: LLMBundle, embd_mdl: LL
         think += rm_query_tags(query_think)
         all_reasoning_steps.append(query_think)
         msg_hisotry.append({"role": "assistant", "content": query_think})
-        search_query = extract_between(query_think, BEGIN_SEARCH_QUERY, END_SEARCH_QUERY)
+        queries = extract_between(query_think, BEGIN_SEARCH_QUERY, END_SEARCH_QUERY)
-        if not search_query:
+        if not queries:
             if ii > 0:
                 break
-            search_query = question
+            queries = [question]
-            txt = f"\n{BEGIN_SEARCH_QUERY}{question}{END_SEARCH_QUERY}\n\n"
-            think += txt
-            msg_hisotry[-1]["content"] += txt
 
+        for search_query in queries:
             logging.info(f"[THINK]Query: {ii}. {search_query}")
             think += f"\n\n> {ii+1}. {search_query}\n\n"
             yield {"answer": think + "</think>", "reference": {}, "audio_binary": None}

rag/nlp/__init__.py

@@ -17,7 +17,6 @@
 import logging
 import random
 from collections import Counter
-from typing import Optional
 
 from rag.utils import num_tokens_from_string
 from . import rag_tokenizer
@@ -604,9 +603,6 @@ def naive_merge_docx(sections, chunk_token_num=128, delimiter="\n。；！？"):
     return cks, images
 
 
-def extract_between(text: str, start_tag: str, end_tag: str) -> Optional[str]:
+def extract_between(text: str, start_tag: str, end_tag: str) -> list[str]:
     pattern = re.escape(start_tag) + r"(.*?)" + re.escape(end_tag)
-    matches = re.findall(pattern, text, flags=re.DOTALL)
+    return re.findall(pattern, text, flags=re.DOTALL)
-    if matches:
-        return matches[-1].strip()
-    return None