注意：本实验在百度平台的Al Studio运行

1.什么是文本分类

文本分类就是根据文本内容将文本划分到不同类别，例如新闻系统中，每篇新闻报道会划归到不同的类别。

2.文本分类的应用

• 内容分类（新闻分类）

• 邮件过滤（例如垃圾邮件过滤）

• 用户分类（如商城消费级别、喜好）

• 评论、文章、对话的情感分类（正面、负面、中性）

3.文本分类案例

• 任务：建立文本分类模型，并对模型进行训练、评估，从而实现对中文新闻摘要类别正确划分

• 数据集：从网站上爬取56821条数据中文新闻摘要，包含10种类别，国际、文化、娱乐、体育、财经、汽车、教育、科技、房产、证券，各类别样本数量如下表所示：

• 模型选择：

• 步骤：

• 代码

  【预处理部分】

  ########################### 数据预处理　#########################
  import os
  from multiprocessing import cpu_count
  import numpy as np
  import paddle
  import paddle.fluid as fluid
  
  #　定义一组公共变量
  data_root = "data/" # 数据集所在目录
  data_file = "news_classify_data.txt" # 原始数据集
  train_file = "train.txt" # 训练集文件
  test_file = "test.txt" # 测试集文件
  dict_file = "dict_txt.txt" # 字典文件(存放字和编码映射关系)
  
  data_file_path = data_root + data_file # 数据集完整路径
  train_file_path = data_root + train_file # 训练集文件完整路径
  test_file_path = data_root + test_file # 测试集文件完整路径
  dict_file_path = data_root + dict_file # 字典文件完整路径
  
  # 取出样本中所有字，对每个字进行编码，将编码结果存入字典文件
  def create_dict():
      dict_set = set() # 集合，用作去重
      with open(data_file_path, "r", encoding="utf-8") as f:
          for line in f.readlines(): # 遍历每行
              line = line.replace("\n", "") # 去除换行符
              tmp_list = line.split("_!_") # 根据分隔符拆分
              title = tmp_list[-1] # 最后一个字段即为标题
              for word in title: # 取出每个字
                  dict_set.add(word)
  
      # 遍历集合，取出每个字进行编号
      dict_txt = {} # 定义字典
      i = 1 # 编码使用的计数器
      for word in dict_set:
          dict_txt[word] = i # 字-编码 键值对添加到字典
          i += 1
  
      dict_txt["<unk>"] = i # 未知字符(在样本中未出现过的字)
  
      # 将字典内容存入文件
      with open(dict_file_path, "w", encoding="utf-8") as f:
          f.write(str(dict_txt))
  
      print("生成字典结束.")
  
  # 传入一个句子，将每个字替换为编码值，和标签一起返回
  def line_encoding(title, dict_txt, label):
      new_line = "" # 编码结果
      for word in title:
          if word in dict_txt: # 在字典中
              code = str(dict_txt[word]) # 取出编码值
          else: # 不在字典中
              code = str(dict_txt["<unk>"]) # 取未知字符编码值
          new_line = new_line + code + "," # 追加到字符串后面
      new_line = new_line[:-1] # 去掉最后一个多余的逗号
      new_line = new_line + "\t" + label + "\n" # 追加标签值
      return new_line
  
  # 读取原始样本，取出标题部分进行编码，将编码后的划分测试集/训练集
  def create_train_test_file():
      # 清空训练集/测试集
      with open(train_file_path, "w") as f:
          pass
      with open(test_file_path, "w") as f:
          pass
  
      # 读取字典文件
      with open(dict_file_path, "r", encoding="utf-8") as f_dict:
          dict_txt = eval(f_dict.readlines()[0]) # 读取字典文件第一行，生成字典对象
  
      # 读取原始样本
      with open(data_file_path, "r", encoding="utf-8") as f_data:
          lines = f_data.readlines()
  
      i = 0
      for line in lines:
          tmp_list = line.replace("\n", "").split("_!_") # 拆分
          title = tmp_list[3] # 标题
          label = tmp_list[1] # 类别
          new_line = line_encoding(title, dict_txt, label) # 对标题编码
  
          if i % 10 == 0: # 写入测试集
              with open(test_file_path, "a", encoding="utf-8") as f:
                  f.write(new_line)
          else: # 写入训练集
              with open(train_file_path, "a", encoding="utf-8") as f:
                  f.write(new_line)
          i += 1
      print("生成训练集/测试集结束.")
  
  create_dict() # 根据样本生成字典
  create_train_test_file()

  输出：

  生成字典结束.
  生成训练集/测试集结束.

  【模型定义与训练】

  paddle.enable_static()
  # 读取字典文件，返回字典长度
  def get_dict_len(dict_path):
      with open(dict_path, "r", encoding="utf-8") as f:
        dict_txt = eval(f.readlines()[0])
      return len(dict_txt.keys())
  
  def data_mapper(sample):
      data, label = sample # 赋值到变量
      val = [int(w) for w in data.split(",")] # 将编码值转换位数字(从文件读取为字符串)
      return val, int(label)
  
  def train_reader(train_file_path): # 训练集读取器
      def reader():
          with open(train_file_path, "r") as f:
              lines = f.readlines()
              np.random.shuffle(lines) # 随机化处理
              for line in lines:
                  data, label = line.split("\t") # 拆分
                  yield data, label
      return paddle.reader.xmap_readers(data_mapper, reader, cpu_count(), 1024)
  
  def test_reader(test_file_path): # 训练集读取器
      def reader():
          with open(test_file_path, "r") as f:
              lines = f.readlines()
              
              for line in lines:
                  data, label = line.split("\t") # 拆分
                  yield data, label
      return paddle.reader.xmap_readers(data_mapper, reader, cpu_count(), 1024)
  
  # 定义网络
  def Text_CNN(data, dict_dim, class_dim=10, emb_dim=128,
               hid_dim=128, hid_dim2=128):
      """
      定义TextCNN模型
      :param data:　输入
      :param dict_dim:　词典大小(词语总的数量)
      :param class_dim:　分类的数量
      :param emb_dim: 词嵌入长度
      :param hid_dim:　第一个卷基层卷积核数量
      :param hid_dim2:　第二个卷基层卷积核数量
      :return:　模型预测结果
      """
      # embedding层
      emb = fluid.layers.embedding(input=data, size=[dict_dim, emb_dim])
      # 并列两个卷积/池化层
      conv1 = fluid.nets.sequence_conv_pool(input=emb, # 输入(词嵌入层输出)
                                            num_filters=hid_dim,#　卷积核数量
                                            filter_size=3,#卷积核大小
                                            act="tanh",#激活函数
                                            pool_type="sqrt")#池化类型
      conv2 = fluid.nets.sequence_conv_pool(input=emb, # 输入(词嵌入层输出)
                                            num_filters=hid_dim2,#　卷积核数量
                                            filter_size=4,#卷积核大小
                                            act="tanh",#激活函数
                                            pool_type="sqrt")#池化类型
      # fc
      output = fluid.layers.fc(input=[conv1, conv2], # 输入
                               size=class_dim,#输出值个数
                               act="softmax")#激活函数
      return output
  
  # 定义占位符张量
  words = fluid.layers.data(name="words",
                            shape=[1],
                            dtype="int64",
                            lod_level=1) # LOD张量用来表示变长数据
  label = fluid.layers.data(name="label",
                            shape=[1],
                            dtype="int64")
  dict_dim = get_dict_len(dict_file_path) # 获取字典长度
  # 调用模型函数
  model = Text_CNN(words, dict_dim)
  # 损失函数
  cost = fluid.layers.cross_entropy(input=model, label=label)
  avg_cost = fluid.layers.mean(cost)
  # 优化器
  optimizer = fluid.optimizer.Adam(learning_rate=0.0001)
  optimizer.minimize(avg_cost)
  # 准确率
  accuracy = fluid.layers.accuracy(input=model, label=label)
  
  # 执行器
  place = fluid.CUDAPlace(0)
  exe = fluid.Executor(place)
  exe.run(fluid.default_startup_program())
  
  # reader
  ## 训练集reader
  tr_reader = train_reader(train_file_path)
  batch_train_reader = paddle.batch(tr_reader, batch_size=128)
  ## 测试集reader
  ts_reader = test_reader(test_file_path)
  batch_test_reader = paddle.batch(ts_reader, batch_size=128)
  
  # feeder
  feeder = fluid.DataFeeder(place=place, feed_list=[words, label])
  
  # 开始训练
  for epoch in range(80): # 外层循环控制训练轮次
      for batch_id, data in enumerate(batch_train_reader()): # 内层循环控制批次
          train_cost, train_acc = exe.run(fluid.default_main_program(),#program
                                          feed=feeder.feed(data),#喂入的参数
                                          fetch_list=[avg_cost, accuracy])#返回值
          if batch_id % 100 == 0:
              print("epoch:%d, batch:%d, cost:%f, acc:%f" %
                    (epoch, batch_id, train_cost[0], train_acc[0]))
  
      # 每轮训练结束后进行模型评估
      test_costs_list = [] # 存放测试集损失值
      test_accs_list = [] # 存放测试集准确率
  
      for batch_id, data in enumerate(batch_test_reader()):
          test_cost, test_acc = exe.run(fluid.default_main_program(), 
                                        feed=feeder.feed(data),
                                        fetch_list=[avg_cost, accuracy])
          test_costs_list.append(test_cost[0])
          test_accs_list.append(test_acc[0])
      #　计算所有批次损失值/准确率均值
      avg_test_cost = sum(test_costs_list) / len(test_costs_list)
      avg_test_acc = sum(test_accs_list) / len(test_accs_list)
      print("epoch:%d, test_cost:%f, test_acc:%f" %
            (epoch, avg_test_cost, avg_test_acc))
  
  # 训练结束，保存模型
  model_save_dir = "model/"
  if not os.path.exists(model_save_dir):
      os.makedirs(model_save_dir)
  fluid.io.save_inference_model(model_save_dir, # 保存路径
                                feeded_var_names=[words.name],# 使用时传入参数名称
                                target_vars=[model],#预测结果
                                executor=exe)#执行器
  print("模型保存成功.")

  输出

  W0301 17:44:27.392561   134 gpu\_resources.cc:61\] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 12.0, Runtime API Version: 11.2
  W0301 17:44:27.396858   134 gpu\_resources.cc:91\] device: 0, cuDNN Version: 8.2.
  
  epoch:0, batch:0, cost:2.311182, acc:0.062500
  epoch:0, batch:100, cost:2.063974, acc:0.398438
  epoch:0, batch:200, cost:1.564050, acc:0.531250
  epoch:0, batch:300, cost:1.113311, acc:0.726562
  epoch:0, test\_cost:0.984438, test\_acc:0.699619
  epoch:1, batch:0, cost:0.904973, acc:0.703125
  epoch:1, batch:100, cost:1.056342, acc:0.648438
  epoch:1, batch:200, cost:0.757222, acc:0.773438
  epoch:1, batch:300, cost:0.768483, acc:0.757812
  epoch:1, test\_cost:0.743878, test\_acc:0.760737
  epoch:2, batch:0, cost:0.799179, acc:0.773438
  epoch:2, batch:100, cost:0.684252, acc:0.796875
  epoch:2, batch:200, cost:0.700367, acc:0.773438
  epoch:2, batch:300, cost:0.658518, acc:0.820312
  epoch:2, test\_cost:0.662426, test\_acc:0.787224
  epoch:3, batch:0, cost:0.732371, acc:0.757812
  epoch:3, batch:100, cost:0.621147, acc:0.804688
  epoch:3, batch:200, cost:0.592039, acc:0.820312
  epoch:3, batch:300, cost:0.566666, acc:0.835938
  epoch:3, test\_cost:0.613746, test\_acc:0.799462
  epoch:4, batch:0, cost:0.477513, acc:0.835938
  epoch:4, batch:100, cost:0.510523, acc:0.820312
  epoch:4, batch:200, cost:0.635614, acc:0.765625
  epoch:4, batch:300, cost:0.487310, acc:0.820312
  epoch:4, test\_cost:0.579772, test\_acc:0.810488
  epoch:5, batch:0, cost:0.564993, acc:0.796875
  epoch:5, batch:100, cost:0.406841, acc:0.882812
  epoch:5, batch:200, cost:0.444392, acc:0.867188
  epoch:5, batch:300, cost:0.729928, acc:0.765625
  epoch:5, test\_cost:0.549828, test\_acc:0.820210
  epoch:6, batch:0, cost:0.676264, acc:0.757812
  epoch:6, batch:100, cost:0.596377, acc:0.859375
  epoch:6, batch:200, cost:0.532443, acc:0.773438
  epoch:6, batch:300, cost:0.534213, acc:0.820312
  epoch:6, test\_cost:0.529354, test\_acc:0.824554
  epoch:7, batch:0, cost:0.426611, acc:0.867188
  epoch:7, batch:100, cost:0.474698, acc:0.851562
  epoch:7, batch:200, cost:0.529180, acc:0.851562
  epoch:7, batch:300, cost:0.512545, acc:0.820312
  epoch:7, test\_cost:0.510936, test\_acc:0.830113
  epoch:8, batch:0, cost:0.559594, acc:0.859375
  epoch:8, batch:100, cost:0.670438, acc:0.812500
  epoch:8, batch:200, cost:0.519166, acc:0.796875
  epoch:8, batch:300, cost:0.487624, acc:0.835938
  epoch:8, test\_cost:0.495487, test\_acc:0.831584
  epoch:9, batch:0, cost:0.523822, acc:0.875000
  epoch:9, batch:100, cost:0.604066, acc:0.812500
  epoch:9, batch:200, cost:0.420712, acc:0.835938
  epoch:9, batch:300, cost:0.490420, acc:0.867188
  epoch:9, test\_cost:0.480619, test\_acc:0.838099
  epoch:10, batch:0, cost:0.337900, acc:0.875000
  epoch:10, batch:100, cost:0.513283, acc:0.820312
  epoch:10, batch:200, cost:0.471449, acc:0.835938
  epoch:10, batch:300, cost:0.387642, acc:0.859375
  epoch:10, test\_cost:0.470769, test\_acc:0.842698
  epoch:11, batch:0, cost:0.448336, acc:0.835938
  epoch:11, batch:100, cost:0.440521, acc:0.843750
  epoch:11, batch:200, cost:0.436573, acc:0.875000
  epoch:11, batch:300, cost:0.694043, acc:0.781250
  epoch:11, test\_cost:0.460047, test\_acc:0.846606
  epoch:12, batch:0, cost:0.481746, acc:0.882812
  epoch:12, batch:100, cost:0.552047, acc:0.804688
  epoch:12, batch:200, cost:0.386960, acc:0.882812
  epoch:12, batch:300, cost:0.341771, acc:0.882812
  epoch:12, test\_cost:0.450299, test\_acc:0.850769
  epoch:13, batch:0, cost:0.547187, acc:0.859375
  epoch:13, batch:100, cost:0.356574, acc:0.875000
  epoch:13, batch:200, cost:0.379301, acc:0.867188
  epoch:13, batch:300, cost:0.374310, acc:0.898438
  epoch:13, test\_cost:0.440953, test\_acc:0.856413
  epoch:14, batch:0, cost:0.689324, acc:0.843750
  epoch:14, batch:100, cost:0.493557, acc:0.851562
  epoch:14, batch:200, cost:0.332372, acc:0.898438
  epoch:14, batch:300, cost:0.462879, acc:0.867188
  epoch:14, test\_cost:0.434527, test\_acc:0.858323
  epoch:15, batch:0, cost:0.539430, acc:0.859375
  epoch:15, batch:100, cost:0.610829, acc:0.828125
  epoch:15, batch:200, cost:0.567205, acc:0.773438
  epoch:15, batch:300, cost:0.448473, acc:0.851562
  epoch:15, test\_cost:0.429657, test\_acc:0.858929
  epoch:16, batch:0, cost:0.275449, acc:0.914062
  epoch:16, batch:100, cost:0.322028, acc:0.890625
  epoch:16, batch:200, cost:0.433238, acc:0.843750
  epoch:16, batch:300, cost:0.435924, acc:0.875000
  epoch:16, test\_cost:0.420777, test\_acc:0.864314
  epoch:17, batch:0, cost:0.463425, acc:0.859375
  epoch:17, batch:100, cost:0.498210, acc:0.890625
  epoch:17, batch:200, cost:0.347210, acc:0.898438
  epoch:17, batch:300, cost:0.375353, acc:0.867188
  epoch:17, test\_cost:0.414920, test\_acc:0.866483
  epoch:18, batch:0, cost:0.371144, acc:0.906250
  epoch:18, batch:100, cost:0.511294, acc:0.828125
  epoch:18, batch:200, cost:0.431728, acc:0.828125
  epoch:18, batch:300, cost:0.505222, acc:0.843750
  epoch:18, test\_cost:0.412018, test\_acc:0.866136
  epoch:19, batch:0, cost:0.417319, acc:0.859375
  epoch:19, batch:100, cost:0.405875, acc:0.867188
  epoch:19, batch:200, cost:0.466319, acc:0.843750
  epoch:19, batch:300, cost:0.524598, acc:0.820312
  epoch:19, test\_cost:0.408254, test\_acc:0.870387
  epoch:20, batch:0, cost:0.278774, acc:0.921875
  epoch:20, batch:100, cost:0.375402, acc:0.875000
  epoch:20, batch:200, cost:0.512493, acc:0.851562
  epoch:20, batch:300, cost:0.352869, acc:0.867188
  epoch:20, test\_cost:0.402862, test\_acc:0.870646
  epoch:21, batch:0, cost:0.328388, acc:0.890625
  epoch:21, batch:100, cost:0.474930, acc:0.843750
  epoch:21, batch:200, cost:0.279459, acc:0.898438
  epoch:21, batch:300, cost:0.480916, acc:0.843750
  epoch:21, test\_cost:0.398193, test\_acc:0.870476
  epoch:22, batch:0, cost:0.360476, acc:0.914062
  epoch:22, batch:100, cost:0.399123, acc:0.867188
  epoch:22, batch:200, cost:0.330940, acc:0.898438
  epoch:22, batch:300, cost:0.449070, acc:0.851562
  epoch:22, test\_cost:0.396272, test\_acc:0.872644
  epoch:23, batch:0, cost:0.311765, acc:0.882812
  epoch:23, batch:100, cost:0.430598, acc:0.859375
  epoch:23, batch:200, cost:0.371466, acc:0.867188
  epoch:23, batch:300, cost:0.497460, acc:0.859375
  epoch:23, test\_cost:0.391935, test\_acc:0.874990
  epoch:24, batch:0, cost:0.278461, acc:0.921875
  epoch:24, batch:100, cost:0.384332, acc:0.867188
  epoch:24, batch:200, cost:0.687089, acc:0.804688
  epoch:24, batch:300, cost:0.465835, acc:0.835938
  epoch:24, test\_cost:0.386384, test\_acc:0.874905
  epoch:25, batch:0, cost:0.359800, acc:0.914062
  epoch:25, batch:100, cost:0.370942, acc:0.906250
  epoch:25, batch:200, cost:0.343612, acc:0.906250
  epoch:25, batch:300, cost:0.373149, acc:0.859375
  epoch:25, test\_cost:0.385754, test\_acc:0.875249
  epoch:26, batch:0, cost:0.359912, acc:0.859375
  epoch:26, batch:100, cost:0.299233, acc:0.906250
  epoch:26, batch:200, cost:0.321898, acc:0.882812
  epoch:26, batch:300, cost:0.506139, acc:0.820312
  epoch:26, test\_cost:0.382092, test\_acc:0.877597
  epoch:27, batch:0, cost:0.438806, acc:0.882812
  epoch:27, batch:100, cost:0.351698, acc:0.867188
  epoch:27, batch:200, cost:0.413263, acc:0.875000
  epoch:27, batch:300, cost:0.327677, acc:0.875000
  epoch:27, test\_cost:0.379122, test\_acc:0.880460
  epoch:28, batch:0, cost:0.329184, acc:0.921875
  epoch:28, batch:100, cost:0.489258, acc:0.882812
  epoch:28, batch:200, cost:0.375317, acc:0.890625
  epoch:28, batch:300, cost:0.355702, acc:0.859375
  epoch:28, test\_cost:0.377964, test\_acc:0.881066
  epoch:29, batch:0, cost:0.360147, acc:0.882812
  epoch:29, batch:100, cost:0.361545, acc:0.906250
  epoch:29, batch:200, cost:0.535644, acc:0.812500
  epoch:29, batch:300, cost:0.463827, acc:0.789062
  epoch:29, test\_cost:0.374992, test\_acc:0.879156
  epoch:30, batch:0, cost:0.386321, acc:0.843750
  epoch:30, batch:100, cost:0.450116, acc:0.851562
  epoch:30, batch:200, cost:0.380319, acc:0.867188
  epoch:30, batch:300, cost:0.357393, acc:0.914062
  epoch:30, test\_cost:0.372232, test\_acc:0.880198
  epoch:31, batch:0, cost:0.338851, acc:0.882812
  epoch:31, batch:100, cost:0.418707, acc:0.890625
  epoch:31, batch:200, cost:0.349568, acc:0.875000
  epoch:31, batch:300, cost:0.414638, acc:0.882812
  epoch:31, test\_cost:0.373127, test\_acc:0.879245
  epoch:32, batch:0, cost:0.278832, acc:0.906250
  epoch:32, batch:100, cost:0.538143, acc:0.851562
  epoch:32, batch:200, cost:0.418359, acc:0.890625
  epoch:32, batch:300, cost:0.510367, acc:0.875000
  epoch:32, test\_cost:0.370239, test\_acc:0.880896
  epoch:33, batch:0, cost:0.410598, acc:0.835938
  epoch:33, batch:100, cost:0.295002, acc:0.906250
  epoch:33, batch:200, cost:0.430560, acc:0.828125
  epoch:33, batch:300, cost:0.417476, acc:0.859375
  epoch:33, test\_cost:0.367410, test\_acc:0.881155
  epoch:34, batch:0, cost:0.337740, acc:0.937500
  epoch:34, batch:100, cost:0.304080, acc:0.906250
  epoch:34, batch:200, cost:0.359049, acc:0.890625
  epoch:34, batch:300, cost:0.373999, acc:0.890625
  epoch:34, test\_cost:0.367002, test\_acc:0.880113
  epoch:35, batch:0, cost:0.411581, acc:0.898438
  epoch:35, batch:100, cost:0.400797, acc:0.851562
  epoch:35, batch:200, cost:0.482271, acc:0.828125
  epoch:35, batch:300, cost:0.340450, acc:0.890625
  epoch:35, test\_cost:0.363663, test\_acc:0.883068
  epoch:36, batch:0, cost:0.338912, acc:0.875000
  epoch:36, batch:100, cost:0.416916, acc:0.867188
  epoch:36, batch:200, cost:0.313621, acc:0.882812
  epoch:36, batch:300, cost:0.677497, acc:0.796875
  epoch:36, test\_cost:0.361819, test\_acc:0.882983
  epoch:37, batch:0, cost:0.329249, acc:0.867188
  epoch:37, batch:100, cost:0.375915, acc:0.890625
  epoch:37, batch:200, cost:0.290267, acc:0.906250
  epoch:37, batch:300, cost:0.388264, acc:0.859375
  epoch:37, test\_cost:0.363713, test\_acc:0.880025
  epoch:38, batch:0, cost:0.452093, acc:0.875000
  epoch:38, batch:100, cost:0.237014, acc:0.898438
  epoch:38, batch:200, cost:0.334976, acc:0.898438
  epoch:38, batch:300, cost:0.386618, acc:0.875000
  epoch:38, test\_cost:0.357681, test\_acc:0.884889
  epoch:39, batch:0, cost:0.397014, acc:0.867188
  epoch:39, batch:100, cost:0.387132, acc:0.882812
  epoch:39, batch:200, cost:0.262646, acc:0.921875
  epoch:39, batch:300, cost:0.295718, acc:0.906250
  epoch:39, test\_cost:0.358814, test\_acc:0.884542
  epoch:40, batch:0, cost:0.336061, acc:0.875000
  epoch:40, batch:100, cost:0.393282, acc:0.867188
  epoch:40, batch:200, cost:0.453071, acc:0.867188
  epoch:40, batch:300, cost:0.276213, acc:0.921875
  epoch:40, test\_cost:0.355846, test\_acc:0.886278
  epoch:41, batch:0, cost:0.362588, acc:0.867188
  epoch:41, batch:100, cost:0.293396, acc:0.914062
  epoch:41, batch:200, cost:0.351766, acc:0.890625
  epoch:41, batch:300, cost:0.437711, acc:0.820312
  epoch:41, test\_cost:0.356017, test\_acc:0.886799
  epoch:42, batch:0, cost:0.431722, acc:0.843750
  epoch:42, batch:100, cost:0.296809, acc:0.914062
  epoch:42, batch:200, cost:0.300333, acc:0.898438
  epoch:42, batch:300, cost:0.392034, acc:0.859375
  epoch:42, test\_cost:0.354504, test\_acc:0.885580
  epoch:43, batch:0, cost:0.237395, acc:0.945312
  epoch:43, batch:100, cost:0.274653, acc:0.914062
  epoch:43, batch:200, cost:0.320165, acc:0.898438
  epoch:43, batch:300, cost:0.233366, acc:0.937500
  epoch:43, test\_cost:0.352862, test\_acc:0.885410
  epoch:44, batch:0, cost:0.309431, acc:0.953125
  epoch:44, batch:100, cost:0.371803, acc:0.843750
  epoch:44, batch:200, cost:0.309721, acc:0.898438
  epoch:44, batch:300, cost:0.330030, acc:0.898438
  epoch:44, test\_cost:0.348967, test\_acc:0.888017
  epoch:45, batch:0, cost:0.382172, acc:0.890625
  epoch:45, batch:100, cost:0.292855, acc:0.929688
  epoch:45, batch:200, cost:0.445127, acc:0.898438
  epoch:45, batch:300, cost:0.365554, acc:0.890625
  epoch:45, test\_cost:0.352218, test\_acc:0.883932
  epoch:46, batch:0, cost:0.424743, acc:0.898438
  epoch:46, batch:100, cost:0.382699, acc:0.859375
  epoch:46, batch:200, cost:0.319472, acc:0.914062
  epoch:46, batch:300, cost:0.414162, acc:0.859375
  epoch:46, test\_cost:0.349987, test\_acc:0.885498
  epoch:47, batch:0, cost:0.304131, acc:0.890625
  epoch:47, batch:100, cost:0.386861, acc:0.890625
  epoch:47, batch:200, cost:0.608894, acc:0.820312
  epoch:47, batch:300, cost:0.281832, acc:0.898438
  epoch:47, test\_cost:0.349286, test\_acc:0.888276
  epoch:48, batch:0, cost:0.406423, acc:0.882812
  epoch:48, batch:100, cost:0.398680, acc:0.898438
  epoch:48, batch:200, cost:0.291706, acc:0.914062
  epoch:48, batch:300, cost:0.358105, acc:0.875000
  epoch:48, test\_cost:0.348130, test\_acc:0.888361
  epoch:49, batch:0, cost:0.284720, acc:0.914062
  epoch:49, batch:100, cost:0.341173, acc:0.898438
  epoch:49, batch:200, cost:0.341595, acc:0.859375
  epoch:49, batch:300, cost:0.442754, acc:0.820312
  epoch:49, test\_cost:0.347218, test\_acc:0.886012
  epoch:50, batch:0, cost:0.311721, acc:0.906250
  epoch:50, batch:100, cost:0.326822, acc:0.875000
  epoch:50, batch:200, cost:0.331799, acc:0.898438
  epoch:50, batch:300, cost:0.426647, acc:0.851562
  epoch:50, test\_cost:0.347288, test\_acc:0.888535
  epoch:51, batch:0, cost:0.389481, acc:0.867188
  epoch:51, batch:100, cost:0.289127, acc:0.906250
  epoch:51, batch:200, cost:0.328051, acc:0.929688
  epoch:51, batch:300, cost:0.426396, acc:0.890625
  epoch:51, test\_cost:0.344246, test\_acc:0.889839
  epoch:52, batch:0, cost:0.288156, acc:0.906250
  epoch:52, batch:100, cost:0.298805, acc:0.906250
  epoch:52, batch:200, cost:0.371176, acc:0.921875
  epoch:52, batch:300, cost:0.389306, acc:0.875000
  epoch:52, test\_cost:0.345692, test\_acc:0.891224
  epoch:53, batch:0, cost:0.425932, acc:0.890625
  epoch:53, batch:100, cost:0.415528, acc:0.882812
  epoch:53, batch:200, cost:0.434767, acc:0.867188
  epoch:53, batch:300, cost:0.331441, acc:0.914062
  epoch:53, test\_cost:0.340924, test\_acc:0.890101
  epoch:54, batch:0, cost:0.260270, acc:0.906250
  epoch:54, batch:100, cost:0.305412, acc:0.898438
  epoch:54, batch:200, cost:0.330370, acc:0.906250
  epoch:54, batch:300, cost:0.334084, acc:0.898438
  epoch:54, test\_cost:0.341799, test\_acc:0.892010
  epoch:55, batch:0, cost:0.239946, acc:0.937500
  epoch:55, batch:100, cost:0.510334, acc:0.898438
  epoch:55, batch:200, cost:0.331789, acc:0.898438
  epoch:55, batch:300, cost:0.273344, acc:0.898438
  epoch:55, test\_cost:0.341348, test\_acc:0.889403
  epoch:56, batch:0, cost:0.288282, acc:0.914062
  epoch:56, batch:100, cost:0.384843, acc:0.898438
  epoch:56, batch:200, cost:0.391903, acc:0.867188
  epoch:56, batch:300, cost:0.352458, acc:0.882812
  epoch:56, test\_cost:0.338860, test\_acc:0.891054
  epoch:57, batch:0, cost:0.434810, acc:0.828125
  epoch:57, batch:100, cost:0.257800, acc:0.953125
  epoch:57, batch:200, cost:0.283473, acc:0.921875
  epoch:57, batch:300, cost:0.337173, acc:0.867188
  epoch:57, test\_cost:0.339060, test\_acc:0.891575
  epoch:58, batch:0, cost:0.240891, acc:0.898438
  epoch:58, batch:100, cost:0.390225, acc:0.875000
  epoch:58, batch:200, cost:0.393483, acc:0.843750
  epoch:58, batch:300, cost:0.289487, acc:0.890625
  epoch:58, test\_cost:0.337302, test\_acc:0.892269
  epoch:59, batch:0, cost:0.210337, acc:0.960938
  epoch:59, batch:100, cost:0.423231, acc:0.867188
  epoch:59, batch:200, cost:0.319490, acc:0.921875
  epoch:59, batch:300, cost:0.451494, acc:0.859375
  epoch:59, test\_cost:0.336483, test\_acc:0.893137
  epoch:60, batch:0, cost:0.231775, acc:0.937500
  epoch:60, batch:100, cost:0.295306, acc:0.906250
  epoch:60, batch:200, cost:0.378960, acc:0.859375
  epoch:60, batch:300, cost:0.350808, acc:0.843750
  epoch:60, test\_cost:0.335058, test\_acc:0.894267
  epoch:61, batch:0, cost:0.440865, acc:0.867188
  epoch:61, batch:100, cost:0.270725, acc:0.882812
  epoch:61, batch:200, cost:0.398181, acc:0.851562
  epoch:61, batch:300, cost:0.363882, acc:0.921875
  epoch:61, test\_cost:0.336761, test\_acc:0.892875
  epoch:62, batch:0, cost:0.321757, acc:0.898438
  epoch:62, batch:100, cost:0.330311, acc:0.890625
  epoch:62, batch:200, cost:0.406124, acc:0.851562
  epoch:62, batch:300, cost:0.275819, acc:0.898438
  epoch:62, test\_cost:0.342463, test\_acc:0.891565
  epoch:63, batch:0, cost:0.321822, acc:0.898438
  epoch:63, batch:100, cost:0.322195, acc:0.882812
  epoch:63, batch:200, cost:0.432605, acc:0.882812
  epoch:63, batch:300, cost:0.377368, acc:0.898438
  epoch:63, test\_cost:0.333785, test\_acc:0.895221
  epoch:64, batch:0, cost:0.247617, acc:0.882812
  epoch:64, batch:100, cost:0.231372, acc:0.921875
  epoch:64, batch:200, cost:0.336805, acc:0.867188
  epoch:64, batch:300, cost:0.274635, acc:0.898438
  epoch:64, test\_cost:0.332033, test\_acc:0.894179
  epoch:65, batch:0, cost:0.241076, acc:0.906250
  epoch:65, batch:100, cost:0.377462, acc:0.906250
  epoch:65, batch:200, cost:0.297226, acc:0.882812
  epoch:65, batch:300, cost:0.440397, acc:0.867188
  epoch:65, test\_cost:0.330794, test\_acc:0.897045
  epoch:66, batch:0, cost:0.266126, acc:0.898438
  epoch:66, batch:100, cost:0.390715, acc:0.859375
  epoch:66, batch:200, cost:0.292437, acc:0.914062
  epoch:66, batch:300, cost:0.395078, acc:0.867188
  epoch:66, test\_cost:0.330902, test\_acc:0.895221
  epoch:67, batch:0, cost:0.301438, acc:0.929688
  epoch:67, batch:100, cost:0.388324, acc:0.898438
  epoch:67, batch:200, cost:0.439915, acc:0.890625
  epoch:67, batch:300, cost:0.310547, acc:0.867188
  epoch:67, test\_cost:0.330386, test\_acc:0.896521
  epoch:68, batch:0, cost:0.243119, acc:0.929688
  epoch:68, batch:100, cost:0.447522, acc:0.875000
  epoch:68, batch:200, cost:0.470691, acc:0.882812
  epoch:68, batch:300, cost:0.296465, acc:0.882812
  epoch:68, test\_cost:0.326098, test\_acc:0.896266
  epoch:69, batch:0, cost:0.260604, acc:0.898438
  epoch:69, batch:100, cost:0.417193, acc:0.882812
  epoch:69, batch:200, cost:0.483119, acc:0.835938
  epoch:69, batch:300, cost:0.405713, acc:0.875000
  epoch:69, test\_cost:0.328661, test\_acc:0.896957
  epoch:70, batch:0, cost:0.300975, acc:0.882812
  epoch:70, batch:100, cost:0.199427, acc:0.945312
  epoch:70, batch:200, cost:0.207260, acc:0.937500
  epoch:70, batch:300, cost:0.199148, acc:0.914062
  epoch:70, test\_cost:0.327545, test\_acc:0.894958
  epoch:71, batch:0, cost:0.281955, acc:0.914062
  epoch:71, batch:100, cost:0.267508, acc:0.914062
  epoch:71, batch:200, cost:0.561389, acc:0.828125
  epoch:71, batch:300, cost:0.377676, acc:0.867188
  epoch:71, test\_cost:0.325637, test\_acc:0.897740
  epoch:72, batch:0, cost:0.348661, acc:0.890625
  epoch:72, batch:100, cost:0.346154, acc:0.898438
  epoch:72, batch:200, cost:0.447819, acc:0.867188
  epoch:72, batch:300, cost:0.342514, acc:0.929688
  epoch:72, test\_cost:0.325294, test\_acc:0.897304
  epoch:73, batch:0, cost:0.223638, acc:0.929688
  epoch:73, batch:100, cost:0.394560, acc:0.859375
  epoch:73, batch:200, cost:0.341260, acc:0.890625
  epoch:73, batch:300, cost:0.283185, acc:0.898438
  epoch:73, test\_cost:0.326340, test\_acc:0.895394
  epoch:74, batch:0, cost:0.371942, acc:0.921875
  epoch:74, batch:100, cost:0.333636, acc:0.882812
  epoch:74, batch:200, cost:0.397030, acc:0.875000
  epoch:74, batch:300, cost:0.392802, acc:0.875000
  epoch:74, test\_cost:0.322571, test\_acc:0.896089
  epoch:75, batch:0, cost:0.275930, acc:0.921875
  epoch:75, batch:100, cost:0.263152, acc:0.914062
  epoch:75, batch:200, cost:0.296550, acc:0.898438
  epoch:75, batch:300, cost:0.402121, acc:0.898438
  epoch:75, test\_cost:0.320611, test\_acc:0.897134
  epoch:76, batch:0, cost:0.279775, acc:0.921875
  epoch:76, batch:100, cost:0.439274, acc:0.843750
  epoch:76, batch:200, cost:0.330266, acc:0.898438
  epoch:76, batch:300, cost:0.418308, acc:0.851562
  epoch:76, test\_cost:0.320242, test\_acc:0.900429
  epoch:77, batch:0, cost:0.320668, acc:0.890625
  epoch:77, batch:100, cost:0.168939, acc:0.960938
  epoch:77, batch:200, cost:0.244379, acc:0.953125
  epoch:77, batch:300, cost:0.621534, acc:0.875000
  epoch:77, test\_cost:0.319756, test\_acc:0.900865
  epoch:78, batch:0, cost:0.284392, acc:0.914062
  epoch:78, batch:100, cost:0.309243, acc:0.890625
  epoch:78, batch:200, cost:0.273962, acc:0.945312
  epoch:78, batch:300, cost:0.311928, acc:0.906250
  epoch:78, test\_cost:0.318491, test\_acc:0.901818
  epoch:79, batch:0, cost:0.242170, acc:0.898438
  epoch:79, batch:100, cost:0.315753, acc:0.875000
  epoch:79, batch:200, cost:0.252874, acc:0.937500
  epoch:79, batch:300, cost:0.447730, acc:0.812500
  epoch:79, test\_cost:0.318828, test\_acc:0.900603
  模型保存成功.

  【推理预测】

  model_save_dir = "model/"
  
  def get_data(sentence): # 将传入的句子根据字典中的值进行编码
      with open(dict_file_path, "r", encoding="utf-8") as f:
          dict_txt = eval(f.readlines()[0])
  
      ret = [] # 编码结果
      keys = dict_txt.keys()
      for w in sentence: # 取出每个字
          if not w in keys: # 字不在字典中
              w = "<unk>"
          ret.append(int(dict_txt[w]))
      return ret
  
  # 执行器
  place = fluid.CPUPlace()
  exe = fluid.Executor(place)
  exe.run(fluid.default_startup_program())
  
  infer_program, feed_names, target_var = \
  fluid.io.load_inference_model(model_save_dir, exe)
  
  texts = [] # 存放待预测句子
  
  data1 = get_data("在获得诺贝尔文学奖7年之后，莫言15日晚间在山西汾阳贾家庄如是说")
  data2 = get_data("综合'今日美国'、《世界日报》等当地媒体报道，芝加哥河滨警察局表示")
  data3 = get_data("中国队2022年冬奥会表现优秀")
  data4 = get_data("中国人民银行今日发布通知，降低准备金率，预计释放4000亿流动性")
  data5 = get_data("10月20日,第六届世界互联网大会正式开幕")
  data6 = get_data("同一户型，为什么高层比低层要贵那么多？")
  data7 = get_data("揭秘A股周涨5%资金动向：追捧2类股，抛售600亿香饽饽")
  data8 = get_data("宋慧乔陷入感染危机，前夫宋仲基不戴口罩露面，身处国外神态轻松")
  data9 = get_data("此盆栽花很好养，花美似牡丹，三季开花，南北都能养，很值得栽培")  # 不属于任何一个类别
  
  texts.append(data1)
  texts.append(data2)
  texts.append(data3)
  texts.append(data4)
  texts.append(data5)
  texts.append(data6)
  texts.append(data7)
  texts.append(data8)
  texts.append(data9)
  
  base_shape = [[len(c) for c in texts]] # 计算每个句子长度
  tensor_words = fluid.create_lod_tensor(texts, base_shape, place)
  result = exe.run(infer_program,
                   feed={feed_names[0]: tensor_words},
                   fetch_list=target_var)
  names = ["文化", "娱乐", "体育", "财经", "房产","汽车", "教育", "科技", "国际", "证券"]
  for r in result[0]:
      idx = np.argmax(r) # 取出最大值的索引
      print("预测结果:", names[idx], " 概率:", r[idx])

  输出

  预测结果: 财经  概率: 0.81440145
  预测结果: 娱乐  概率: 1.0
  预测结果: 财经  概率: 1.0
  预测结果: 汽车  概率: 0.9996093
  预测结果: 文化  概率: 0.9404757
  预测结果: 娱乐  概率: 0.8715788
  预测结果: 房产  概率: 0.9625704
  预测结果: 科技  概率: 0.985617
  预测结果: 房产  概率: 1.0

  文章涉及到的数据资源链接如下：news_classify_data.zip - 蓝奏云文件大小：2.7 M|https://wwt.lanzoum.com/iTF6N1q1rnti