Update gif for readme and add input param to every components (#3145)

### What problem does this PR solve?


### Type of change

- [x] New Feature (non-breaking change which adds functionality)
- [x] Documentation Update

README.md

@@ -69,11 +69,12 @@ data.
 Try our demo at [https://demo.ragflow.io](https://demo.ragflow.io).
 <div align="center" style="margin-top:20px;margin-bottom:20px;">
 <img src="https://github.com/infiniflow/ragflow/assets/7248/2f6baa3e-1092-4f11-866d-36f6a9d075e5" width="1200"/>
-<img src="https://github.com/infiniflow/ragflow/assets/12318111/b083d173-dadc-4ea9-bdeb-180d7df514eb" width="1200"/>
+<img src="https://github.com/user-attachments/assets/504bbbf1-c9f7-4d83-8cc5-e9cb63c26db6" width="1200"/>
 </div>
 
 ## 🔥 Latest Updates
 
+- 2024-11-01 Adds keyword extraction and related question generation to the parsed chunk to improve the accuracy of retrieval.
 - 2024-09-29 Optimizes multi-round conversations.
 - 2024-09-13 Adds search mode for knowledge base Q&A.
 - 2024-09-09 Adds a medical consultant agent template.

README_ja.md

@@ -47,12 +47,13 @@
 デモをお試しください：[https://demo.ragflow.io](https://demo.ragflow.io)。
 <div align="center" style="margin-top:20px;margin-bottom:20px;">
 <img src="https://github.com/infiniflow/ragflow/assets/7248/2f6baa3e-1092-4f11-866d-36f6a9d075e5" width="1200"/>
-<img src="https://github.com/infiniflow/ragflow/assets/12318111/b083d173-dadc-4ea9-bdeb-180d7df514eb" width="1200"/>
+<img src="https://github.com/user-attachments/assets/504bbbf1-c9f7-4d83-8cc5-e9cb63c26db6" width="1200"/>
 </div>
 
 
 ## 🔥 最新情報
 
+- 2024-11-01 再現の精度を向上させるために、解析されたチャンクにキーワード抽出と関連質問の生成を追加しました。
 - 2024-09-29 マルチラウンドダイアログを最適化。
 - 2024-09-13 ナレッジベース Q&A の検索モードを追加しました。
 - 2024-09-09 エージェントに医療相談テンプレートを追加しました。

README_ko.md

@@ -49,12 +49,14 @@
 데모를 [https://demo.ragflow.io](https://demo.ragflow.io)에서 실행해 보세요.
 <div align="center" style="margin-top:20px;margin-bottom:20px;">
 <img src="https://github.com/infiniflow/ragflow/assets/7248/2f6baa3e-1092-4f11-866d-36f6a9d075e5" width="1200"/>
-<img src="https://github.com/infiniflow/ragflow/assets/12318111/b083d173-dadc-4ea9-bdeb-180d7df514eb" width="1200"/>
+<img src="https://github.com/user-attachments/assets/504bbbf1-c9f7-4d83-8cc5-e9cb63c26db6" width="1200"/>
 </div>
 
 
 ## 🔥 업데이트
 
+- 2024-11-01 파싱된 청크에 키워드 추출 및 관련 질문 생성을 추가하여 재현율을 향상시킵니다.
+  
 - 2024-09-29 다단계 대화를 최적화합니다.
   
 - 2024-09-13 지식베이스 Q&A 검색 모드를 추가합니다.

README_zh.md

@@ -47,12 +47,13 @@
 请登录网址 [https://demo.ragflow.io](https://demo.ragflow.io) 试用 demo。
 <div align="center" style="margin-top:20px;margin-bottom:20px;">
 <img src="https://github.com/infiniflow/ragflow/assets/7248/2f6baa3e-1092-4f11-866d-36f6a9d075e5" width="1200"/>
-<img src="https://github.com/infiniflow/ragflow/assets/12318111/b083d173-dadc-4ea9-bdeb-180d7df514eb" width="1200"/>
+<img src="https://github.com/user-attachments/assets/504bbbf1-c9f7-4d83-8cc5-e9cb63c26db6" width="1200"/>
 </div>
 
 
 ## 🔥 近期更新
 
+- 2024-11-01 对解析后的chunk加入关键词抽取和相关问题生成以提高召回的准确度。
 - 2024-09-29 优化多轮对话.
 - 2024-09-13 增加知识库问答搜索模式。
 - 2024-09-09 在 Agent 中加入医疗问诊模板。

agent/component/base.py

@@ -36,6 +36,7 @@ class ComponentParamBase(ABC):
     def __init__(self):
         self.output_var_name = "output"
         self.message_history_window_size = 22
+        self.query = []
 
     def set_name(self, name: str):
         self._name = name
@@ -436,6 +437,16 @@ class ComponentBase(ABC):
         setattr(self._param, self._param.output_var_name, v)
 
     def get_input(self):
+        if self._param.query:
+            outs = []
+            for q in self._param.query:
+                if q["value"]: outs.append(pd.DataFrame([{"content": q["value"]}]))
+                if q["component_id"]: outs.append(self._canvas.get_component(q["component_id"])["obj"].output(allow_partial=False)[1])
+            if outs:
+                df = pd.concat(outs, ignore_index=True)
+                if "content" in df: df = df.drop_duplicates(subset=['content']).reset_index(drop=True)
+                return df
+
         upstream_outs = []
         reversed_cpnts = []
         if len(self._canvas.path) > 1:

agent/component/generate.py

@@ -130,6 +130,7 @@ class Generate(ComponentBase):
 
         msg = self._canvas.get_history(self._param.message_history_window_size)
         _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(chat_mdl.max_length * 0.97))
+        if len(msg) < 2: msg.append({"role": "user", "content": ""})
         ans = chat_mdl.chat(msg[0]["content"], msg[1:], self._param.gen_conf())
 
         if self._param.cite and "content_ltks" in retrieval_res.columns and "vector" in retrieval_res.columns:
@@ -149,6 +150,7 @@ class Generate(ComponentBase):
 
         msg = self._canvas.get_history(self._param.message_history_window_size)
         _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(chat_mdl.max_length * 0.97))
+        if len(msg) < 2: msg.append({"role": "user", "content": ""})
         answer = ""
         for ans in chat_mdl.chat_streamly(msg[0]["content"], msg[1:], self._param.gen_conf()):
             res = {"content": ans, "reference": []}

agent/component/invoke.py

@@ -51,6 +51,9 @@ class Invoke(ComponentBase, ABC):
         for para in self._param.variables:
             if para.get("component_id"):
                 cpn = self._canvas.get_component(para["component_id"])["obj"]
+                if cpn.component_name.lower() == "answer":
+                    args[para["key"]] = self._canvas.get_history(1)[0]["content"]
+                    continue
                 _, out = cpn.output(allow_partial=False)
                 args[para["key"]] = "\n".join(out["content"])
             else:

agent/templates/HR_callout_zh.json

@@ -152,7 +152,8 @@
             "Generate:ToughLawsCheat",
             "Generate:KindCarrotsSit",
             "Generate:DirtyToolsTrain",
-            "Generate:FluffyPillowsGrow"
+            "Generate:FluffyPillowsGrow",
+            "Generate:ProudEarsWorry"
           ]
         },
         "Retrieval:ShaggyRadiosRetire": {
@@ -212,7 +213,9 @@
               "top_p": 0.3
             }
           },
-          "downstream": [],
+          "downstream": [
+            "Answer:TwentyMugsDeny"
+          ],
           "upstream": [
             "categorize:0"
           ]
@@ -331,9 +334,9 @@
               "message_history_window_size": 12,
               "parameters": [
                 {
+                  "component_id": "Retrieval:ColdEelsArrive",
                   "id": "5166a107-e859-4c71-99a2-3a216c775347",
-                  "key": "jd",
-                  "component_id": "Retrieval:ColdEelsArrive"
+                  "key": "jd"
                 }
               ],
               "presence_penalty": 0.4,
@@ -1266,9 +1269,9 @@
                 "parameter": "Precise",
                 "parameters": [
                   {
+                    "component_id": "Retrieval:ColdEelsArrive",
                     "id": "5166a107-e859-4c71-99a2-3a216c775347",
-                    "key": "jd",
-                    "component_id": "Retrieval:ColdEelsArrive"
+                    "key": "jd"
                   }
                 ],
                 "presencePenaltyEnabled": true,
@@ -1541,6 +1544,19 @@
             "target": "Answer:TwentyMugsDeny",
             "targetHandle": "c",
             "type": "buttonEdge"
+          },
+          {
+            "type": "buttonEdge",
+            "markerEnd": "logo",
+            "style": {
+              "strokeWidth": 2,
+              "stroke": "rgb(202 197 245)"
+            },
+            "source": "Generate:ProudEarsWorry",
+            "sourceHandle": "b",
+            "target": "Answer:TwentyMugsDeny",
+            "targetHandle": "c",
+            "id": "reactflow__edge-Generate:ProudEarsWorryb-Answer:TwentyMugsDenyc"
           }
         ]
       },