新增差值阈值

final/ByRules/app.py

@@ -3,8 +3,8 @@ import sys
 from flask import Flask, request, jsonify
 
 from commonUtil import fill_template, fill_template_auto
-from similarity_answer_json_copy import *
-# from similarity_answer_json import *
+# from similarity_answer_json_copy import *
+from similarity_answer_json import *
 from util import *
 import os
 
@@ -28,8 +28,8 @@ app = Flask(__name__)
 
 
 BASE_DIR = os.path.dirname(os.path.abspath(__file__))
-TEMPLATE_FOLDER = os.path.join(BASE_DIR, "templatesJson_copy")
-# TEMPLATE_FOLDER = os.path.join(BASE_DIR, "templatesJson")
+# TEMPLATE_FOLDER = os.path.join(BASE_DIR, "templatesJson_copy")
+TEMPLATE_FOLDER = os.path.join(BASE_DIR, "templatesJson")
 DATA_FOLDER = os.path.join(BASE_DIR, "..", "Json", "json_data")
 
 MAPPING_FOLDER = os.path.join(BASE_DIR, "..", "Json", "sjgxys")
@@ -45,22 +45,22 @@ def process_query_route():
         return jsonify({"error": "Query cannot be empty"}), 400
 
     try:
-        # result = process_query(query, template_dict, TEMPLATE_FOLDER)
-        result = process_query(query, template_dict, TEMPLATE_FOLDER,item_lexicon)
-
-        print("用户问句：", query)
-        print("条件为：",result['conditions'])
-        print("匹配的模板 key：", result['matched_key'])
-        print("最相似的模板：", result['matched_template'])
-        print("相似度分数：", result['similarity_score'])
-        print("类型：", result["type"])
-        print("关键词：", result["keywords"])
-        print("查询字段：", result["target"])
-        print("模型名字", result["name"])
-        print("条件", result["conditions"])
-        print("返回的内容是：", result["content"])
-        print("问句是：", result["query"])
-        print("动作是：", result["play"])
+        result = process_query(query, template_dict, TEMPLATE_FOLDER)
+        # result = process_query(query, template_dict, TEMPLATE_FOLDER,item_lexicon)
+
+        # print("用户问句：", query)
+        # print("条件为：",result['conditions'])
+        # print("匹配的模板 key：", result['matched_key'])
+        # print("最相似的模板：", result['matched_template'])
+        # print("相似度分数：", result['similarity_score'])
+        # print("类型：", result["type"])
+        # print("关键词：", result["keywords"])
+        # print("查询字段：", result["target"])
+        # print("模型名字", result["name"])
+        # print("条件", result["conditions"])
+        # print("返回的内容是：", result["content"])
+        # print("问句是：", result["query"])
+        # print("动作是：", result["play"])
 
         #  如果没有该问题模板
         if result['play'] == '疑问':

final/ByRules/similarity_answer_json.py

@@ -373,6 +373,40 @@ def match_template(query, template_dict, tokenizer):
 
     return matched_key, best_match_sentence, similarity_score
 
+def match_template_all(query, template_dict, tokenizer):
+    """
+    返回所有模板句与 query 的匹配得分列表，格式为 [(key, 模板句, 相似度分数), ...]，按分数降序排列。
+    """
+    templates = []
+    key_map = []
+    for key, sentences in template_dict.items():
+        for s in sentences:
+            templates.append(s)
+            key_map.append(key)
+
+    if not templates:
+        return []
+
+    vectorizer = TfidfVectorizer(tokenizer=tokenizer)
+    tfidf_matrix = vectorizer.fit_transform([query] + templates)
+    cos_sim = cosine_similarity(tfidf_matrix[0:1], tfidf_matrix[1:])[0]
+
+    results = []
+    for idx, score in enumerate(cos_sim):
+        results.append((key_map[idx], templates[idx], score))
+
+    # 按相似度降序排序
+    results.sort(key=lambda x: x[2], reverse=True)
+
+    return results
+
+def match_template_with_time_filter_all(query, template_dict, tokenizer, extract_time_location_func):
+    time_info, _, _, _ = extract_time_location_func(query)
+    candidate_keys = classify_by_time_type(query, time_info)
+    filtered_template_dict = {k: template_dict[k] for k in candidate_keys if k in template_dict}
+    return match_template_all(query, filtered_template_dict, tokenizer)
+
+
 # 根据模板去对应的json文件中找数据
 def load_template_info(matched_key, json_folder):
     """
@@ -405,40 +439,60 @@ def process_query(query, template_dict, json_folder, tokenizer=jieba_tokenizer):
 
     conditions = {}
     # 匹配模板
-    matched_key, best_sentence, score = match_template_with_time_filter(
-        query,
-        template_dict,
-        tokenizer,
-        extract_time_location_func=extract_time_location
-    )
-    # 定义阈值
+    # matched_key, best_sentence, score = match_template_with_time_filter(
+    #     query,
+    #     template_dict,
+    #     tokenizer,
+    #     extract_time_location_func=extract_time_location
+    # )
+
+    # 这里match_template_with_time_filter改为返回所有匹配的列表 [(matched_key, best_sentence, score), ...]
+    all_matches = match_template_with_time_filter_all(query, template_dict, tokenizer,
+                                                      extract_time_location_func=extract_time_location)
+    for idx, (key, sentence, score) in enumerate(all_matches):
+        print(f"排名 {idx + 1}: key={key}, 模板句='{sentence}', 相似度={score:.4f}")
+
+    # 按得分排序（降序）
+    all_matches.sort(key=lambda x: x[2], reverse=True)
+    best_match = all_matches[0]
+    best_score = best_match[2]
+    second_score = all_matches[1][2] if len(all_matches) > 1 else 0
+
+    # 判断阈值
     similarity_threshold = 0.25
-    # ★ 判断相似度阈值
-    if score < similarity_threshold:
+    diff_threshold = 0.05  # 差距阈值，可调
+
+    if best_score < similarity_threshold:
         return {
             "matched_key": None,
             "matched_template": None,
-            "similarity_score": score,
-            "type": None,
-            "keywords": None,
-            "target": None,
-            "name": None,
-            "conditions": conditions,
+            "similarity_score": best_score,
             "content": "您提问的问题目前我还没有掌握",
             "query": query,
-            "play":"疑问"
+            "play": "疑问"
+        }
+
+    if (best_score - second_score) < diff_threshold:
+        # 差距太小，匹配不准确
+        return {
+            "content": "您提问的问题不太准确，我无法理解",
+            "query": query,
+            "play": "疑问",
+            "name": "疑问",
         }
 
+    # 匹配准确，正常返回第一个匹配的模板信息
+    matched_key, best_sentence, score = best_match
+
+    # 下面是你已有的提取条件逻辑
     if time_info:
         ti = time_info[0]
-        # 先判断是否是区间时间（有start_year/end_year等字段）
         if 'start_year' in ti and 'end_year' in ti:
             conditions['start_year'] = ti.get('start_year')
             conditions['start_month'] = ti.get('start_month')
             conditions['end_year'] = ti.get('end_year')
             conditions['end_month'] = ti.get('end_month')
         else:
-            # 单时间点
             if 'year' in ti:
                 conditions['年'] = ti['year']
             if 'month' in ti:
@@ -455,28 +509,19 @@ def process_query(query, template_dict, json_folder, tokenizer=jieba_tokenizer):
     if rank_info2:
         conditions['rank2'] = rank_info2
 
-    # 查询模板json
     template_info = load_template_info(matched_key, json_folder)
-    # 模板的关键词
     keywords = template_info.get("keyword")
-    # 模板中的映射关系
     target = template_info.get("target")
-    # 模板的类型
     type_ = template_info.get("type", "")
-    # 模板的名字
     dataJsonName = template_info.get("dataJsonName", "")
-    # ---------------- 比较类 -----------------
     value_key = template_info.get("value_key", "")
     name_key = template_info.get("name_key", "")
     find_max = template_info.get("find_max")
-    # block名称
     name = template_info.get("name", "")
-    # 输出内容
     content = template_info.get("content", "")
-    # 动作类型
     play = template_info.get("play", "")
-    # 问题序号
     qcode = template_info.get("qcode", "")
+
     return {
         "matched_key": matched_key,
         "matched_template": best_sentence,
@@ -582,7 +627,7 @@ def find_key_recursively(data, target_key):
 # print("动作是：", result["play"])
 
 # query = "当月送出均价最高的是哪个省？？"
-# query = ("2025年送出电量前五名是谁？？")
+# query = ("交易？")
 #
 # json_folder = "templatesJson"
 #
@@ -599,6 +644,7 @@ def find_key_recursively(data, target_key):
 # print("返回的内容是：", result["content"])
 # print("问句是：", result["query"])
 # print("动作是：", result["play"])
+# print("描述：", result["content"])
 
 #
 # type = result["type"]