feat: internet search speed and reranker (#282)

* feat: modify search rephrase prompt

* feat: modify task parser prompt

* feat: add searched log

Author: CaralHsi

src/memos/memories/textual/tree_text_memory/retrieve/bochasearch.py

@@ -218,7 +218,9 @@ def _process_result(
         memory_items = []
         for read_item_i in read_items[0]:
             read_item_i.memory = (
-                f"Title: {title}\nNewsTime: {publish_time}\nSummary: {summary}\n"
+                f"[Outer internet view] Title: {title}\nNewsTime:"
+                f" {publish_time}\nSummary:"
+                f" {summary}\n"
                 f"Content: {read_item_i.memory}"
             )
             read_item_i.metadata.source = "web"

src/memos/memories/textual/tree_text_memory/retrieve/recall.py

@@ -179,7 +179,7 @@ def _vector_recall(
         query_embedding: list[list[float]],
         memory_scope: str,
         top_k: int = 20,
-        max_num: int = 5,
+        max_num: int = 3,
         cube_name: str | None = None,
     ) -> list[TextualMemoryItem]:
         """

src/memos/memories/textual/tree_text_memory/retrieve/searcher.py

@@ -87,6 +87,12 @@ def search(
         self._update_usage_history(final_results, info)
 
         logger.info(f"[SEARCH] Done. Total {len(final_results)} results.")
+        res_results = ""
+        for _num_i, result in enumerate(final_results):
+            res_results += "\n" + (
+                result.id + "|" + result.metadata.memory_type + "|" + result.memory
+            )
+        logger.info(f"[SEARCH] Results. {res_results}")
         return final_results
 
     @timed
@@ -108,9 +114,10 @@ def _parse_task(self, query, info, mode, top_k=5):
             context = list({node["memory"] for node in related_nodes})
 
             # optional: supplement context with internet knowledge
-            if self.internet_retriever:
+            """if self.internet_retriever:
                 extra = self.internet_retriever.retrieve_from_internet(query=query, top_k=3)
                 context.extend(item.memory.partition("\nContent: ")[-1] for item in extra)
+            """
 
         # parse goal using LLM
         parsed_goal = self.task_goal_parser.parse(

src/memos/memories/textual/tree_text_memory/retrieve/utils.py

@@ -8,18 +8,20 @@
 5. Need for internet search: If the user's task instruction only involves objective facts or can be completed without introducing external knowledge, set "internet_search" to False. Otherwise, set it to True.
 6. Memories: Provide 2–5 short semantic expansions or rephrasings of the rephrased/original user task instruction. These are used for improved embedding search coverage. Each should be clear, concise, and meaningful for retrieval.
 
-Task description:
-\"\"\"$task\"\"\"
-
 Former conversation (if any):
 \"\"\"
 $conversation
 \"\"\"
 
+Task description(User Question):
+\"\"\"$task\"\"\"
+
 Context (if any):
 \"\"\"$context\"\"\"
 
-Return strictly in this JSON format:
+Return strictly in this JSON format, note that the
+keys/tags/rephrased_instruction/memories should use the same language as the
+input query:
 {
   "keys": [...],
   "tags": [...],

src/memos/reranker/http_bge.py

@@ -1,6 +1,8 @@
 # memos/reranker/http_bge.py
 from __future__ import annotations
 
+import re
+
 from typing import TYPE_CHECKING
 
 import requests
@@ -11,6 +13,8 @@
 if TYPE_CHECKING:
     from memos.memories.textual.item import TextualMemoryItem
 
+_TAG1 = re.compile(r"^\s*\[[^\]]*\]\s*")
+
 
 class HTTPBGEReranker(BaseReranker):
     """
@@ -43,7 +47,10 @@ def rerank(
         if not graph_results:
             return []
 
-        documents = [getattr(item, "memory", None) for item in graph_results]
+        documents = [
+            (_TAG1.sub("", m) if isinstance((m := getattr(item, "memory", None)), str) else m)
+            for item in graph_results
+        ]
         documents = [d for d in documents if isinstance(d, str) and d]
         if not documents:
             return []