from WxMini.Milvus.Config.MulvusConfig import * from WxMini.Milvus.Utils.MilvusCollectionManager import MilvusCollectionManager from WxMini.Milvus.Utils.MilvusConnectionPool import * from gensim.models import KeyedVectors import jieba import time # 1. 加载预训练的 Word2Vec 模型 model_path = MS_MODEL_PATH model = KeyedVectors.load_word2vec_format(model_path, binary=False, limit=MS_MODEL_LIMIT) print(f"模型加载成功,词向量维度: {model.vector_size}") # 功能:将文本转换为嵌入向量 def text_to_embedding(text): words = jieba.lcut(text) # 使用 jieba 分词 print(f"文本: {text}, 分词结果: {words}") embeddings = [model[word] for word in words if word in model] print(f"有效词向量数量: {len(embeddings)}") if embeddings: avg_embedding = sum(embeddings) / len(embeddings) print(f"生成的平均向量: {avg_embedding[:5]}...") # 打印前 5 维 return avg_embedding else: print("未找到有效词,返回零向量") return [0.0] * model.vector_size # 2. 使用连接池管理 Milvus 连接 milvus_pool = MilvusConnectionPool(host=MS_HOST, port=MS_PORT, max_connections=MS_MAX_CONNECTIONS) # 3. 从连接池中获取一个连接 connection = milvus_pool.get_connection() # 4. 初始化集合管理器 collection_name = MS_COLLECTION_NAME collection_manager = MilvusCollectionManager(collection_name) # 5. 输入一个用户问题 user_input = input("请输入一句话:") # 例如:“我今天心情不太好” model_response = "我没听懂,能再说一遍吗?" # 大模型的固定回复 timestamp = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime()) # 当前时间 session_id = "session_001" # 会话 ID(可以根据需要动态生成) # 6. 将用户问题转换为嵌入向量 user_embedding = text_to_embedding(user_input) # 7. 插入数据,确保字段顺序与集合定义一致 entities = [ [session_id], # session_id [user_input], # user_input [model_response], # model_response [timestamp], # timestamp [user_embedding] # embedding ] collection_manager.insert_data(entities) print("数据插入成功。") # 8. 释放连接 milvus_pool.release_connection(connection) # 9. 关闭连接池 milvus_pool.close()