上传文件至 kg_lab_6.13

6.17 更新对检索工程数据复杂表达式的能力
2025-06-17 17:17:20 +08:00
parent 8a44b9780d
commit fad7c5de4a
3 changed files with 85 additions and 42 deletions
@@ -3,10 +3,20 @@ from vector_lab import intersection_of_three_lists
 from utils import find_target_item, find_target_items, pre_mapping, pre_mapping2
 import json

+# 样例
+# input_str1 = "杆塔总基数是多少？"
+# input_str2 = "单回路长度是多少？"
+# input_str3 = "计算一下角钢塔的塔材装材费"
+# input_str4 = "计算一下土石方总量"
+# input_str5 = "板式塔基的各类基础数量占总塔基数比例是多少？"
+# input_str6 = "基础混凝土总量是多少"
+# input_str7 = "计算一下本体工程机械费"
+# input_str8 = "项目建设技术服务费合计"
+
 # 初始化
 problem_rewrite = Problem_rewrite()

-from utils import extract_concrete_info, extract_query_prefix_list
+from utils import extract_concrete_info, extract_query_prefix_list, split_chinese_bracketed_phrases

 from chains_lab import question_answer, question_answer_calculation

@@ -76,14 +86,16 @@ while True:

        elif isinstance(question, list):
            ques = extract_query_prefix_list(question)
-            ques_info = extract_concrete_info(question)
+            temp = extract_concrete_info(question)
+            ques_info = split_chinese_bracketed_phrases(temp[0])
+            print(ques)

            retriever_info = []
-            for i in ques:
-                response = booway_cypher_chain.invoke(ques)    
-                # generated_cypher = response.get("intermediate_steps")[0]
+            for idx, i in enumerate(ques):
+                response = booway_cypher_chain.invoke(i)    
                temp = response.get("result")
                retriever_info.append(temp)
+        

            retriever_keywords = ques_info[0]
            calculation = ques_info[-1]
@@ -112,7 +112,7 @@ def pre_mapping(keywords, data):
                            if judge_exists(item, data):
                                temp0 = item
                                # temp0 = find_target_items(ceshi["指标描述"]["映射规则"], item, data)
-                                result.append(f"模糊查找一下【{temp0}】，换算规则：【{temp1}】,233")
+                                result.append(f"模糊查找一下【{temp0}】，换算规则：【{temp1}】")
                            else:
                                continue
    
@@ -125,45 +125,39 @@ def pre_mapping(keywords, data):

                return result

-
-def extract_concrete_info(outputs):
+# 
+def extract_query_prefix_list(text_list):
    import re
-    from typing import List
+    pattern = r'^.*?【[^】]*】'
+    return [re.search(pattern, s).group() for s in text_list if re.search(pattern, s)]

-    """
-    从多个句子中提取第一个“【】”作为查找信息，最后一个“【】”作为换算规则，
-    返回格式为：[合并的查找句子, 换算规则]
-    """
-    prefixes = []
-    suffix = ''
-
-    for item in outputs:
-        matches = re.findall(r'【([^】]+)】', item)
-        if len(matches) >= 2:
-            prefixes.append(f"查找一下【{matches[0]}】")
-            # 假设所有换算规则一致，取第一个即可
-            if not suffix:
-                suffix = f'换算规则：【{matches[-1]}】'
-
-    if not prefixes or not suffix:
-        return []
-
-    return ['; '.join(prefixes), suffix]
-
-
-def extract_query_prefix_list(input_list):
+def extract_concrete_info(ceshi):
    import re
-    """
-    输入一个字符串列表，提取每个字符串中符合格式的前缀内容（例如：'查找一下【样式】'）
+    keyword_list = []
+    rule_text = None
+
+    for item in ceshi:
+        # 提取关键词
+        keyword_match = re.search(r'查找一下【(.*?)】', item)
+        if keyword_match:
+            keyword_list.append(keyword_match.group(1))
+        
+        # 提取多行规则，使用 DOTALL 模式使 . 匹配换行符
+        rule_match = re.search(r'换算规则：【(.*?)】', item, re.DOTALL)
+        if rule_match and rule_text is None:
+            rule_text = rule_match.group(1)  # 只取第一个规则内容，假设所有项规则一致
+
+    merged = f"模糊查找一下【{'；'.join(keyword_list)}】，换算规则：【{rule_text}】"
+    return [merged]
+
+def split_chinese_bracketed_phrases(text):
+    import re
+    # 使用正则匹配【...】结构和其前面的标识词
+    pattern = r'[^【]*?【[^】]*】'
+    matches = re.findall(pattern, text)
+    return [match.strip() for match in matches]

-    参数:
-        input_list (list[str]): 包含描述性语句的字符串列表

-    返回:
-        list[str]: 提取出的前缀部分列表（如 '查找一下【大板式】'）
-    """
-    pattern = r'(查找一下【[^】]+】)'
-    return [re.match(pattern, text).group(1) for text in input_list if re.match(pattern, text)]



@@ -1,12 +1,49 @@
 import os
 from langchain_community.vectorstores import FAISS
 from langchain_huggingface import HuggingFaceEmbeddings
+from langchain.embeddings.base import Embeddings
+from openai import OpenAI
+import requests
+import httpx
+import logging
+
+class SiliconFlowEmbeddings(Embeddings):
+    """SiliconFlow嵌入模型封装"""
+    def __init__(self, api_key: str, model: str = "bge-m3"):
+        self.api_key = api_key
+        self.model = model
+        self.url = "http://10.1.16.39:9995/v1/embeddings"
+        self.headers = {
+            "Authorization": f"Bearer {self.api_key}",
+            "Content-Type": "application/json"
+        }
+
+    def _embed(self, input):
+        payload = {
+            "model": self.model,
+            "input": input,
+            "encoding_format": "float"
+        }
+        response = requests.post(self.url, json=payload, headers=self.headers)
+        response.raise_for_status()
+        data = response.json()
+        return [item["embedding"] for item in data["data"]]
+
+    def embed_documents(self, texts):
+        return self._embed(texts)
+
+    def embed_query(self, text):
+        return self._embed([text])[0]
+
+
+# embeddings = Embedding(url="http://10.1.16.39:9995/v1", api_key="xxx", model_name="bge-m3")
+embeddings  = SiliconFlowEmbeddings(api_key="xxx")

 with open("./data/data.txt", 'r', encoding='utf-8') as file:
    txt_list = [line.strip() for line in file]

-embedding_path = "/data/Z_LLM_data/Embed_data/bge-m3"
-embeddings = HuggingFaceEmbeddings(model_name=embedding_path)
+# embedding_path = "/data/Z_LLM_data/Embed_data/bge-m3"
+# embeddings = HuggingFaceEmbeddings(model_name=embedding_path)

 faiss_archived = "./data/faiss_data/data"
 vectorstore_txt_faiss = FAISS.from_texts(txt_list, embeddings)