### PyTorch 实现卷积神经网络 (CNN) 对 THUCNews 新闻文本分类 以下是基于 PyTorch 使用卷积神经网络 (CNN) 进行 THUCNews 数据集上的新闻文本分类的教程和代码示例： #### 1. 数据预处理 在构建模型之前，需要对数据进行清洗、分词以及向量化操作。通常会使用 `jieba` 或其他中文分词工具完成分词工作，并通过嵌入层将词语映射为固定维度的向量。 ```python import torch from torch.utils.data import Dataset, DataLoader from sklearn.model_selection import train_test_split import jieba class NewsDataset(Dataset): def __init__(self, texts, labels, word_to_idx, max_len=500): self.texts = texts self.labels = labels self.word_to_idx = word_to_idx self.max_len = max_len def __len__(self): return len(self.texts) def __getitem__(self, idx): text = self.texts[idx][:self.max_len] label = self.labels[idx] # 将文本转换为索引列表 token_ids = [self.word_to_idx.get(word, 1) for word in jieba.lcut(text)] # 未知单词用1表示 padding_length = self.max_len - len(token_ids) if padding_length > 0: token_ids += [0] * padding_length # 填充至最大长度 else: token_ids = token_ids[:self.max_len] return { 'input_ids': torch.tensor(token_ids, dtype=torch.long), 'labels': torch.tensor(label, dtype=torch.long) } ``` #### 2. 构建 CNN 模型 定义一个简单的 TextCNN 模型，该模型由多个一维卷积核组成，用于捕获不同大小的 n-gram 特征[^2]。 ```python import torch.nn as nn class TextCNN(nn.Module): def __init__(self, vocab_size, embedding_dim, num_filters, filter_sizes, output_dim, dropout_prob): super(TextCNN, self).__init__() self.embedding = nn.Embedding(vocab_size, embedding_dim) self.convs = nn.ModuleList([ nn.Conv1d(in_channels=embedding_dim, out_channels=num_filters, kernel_size=fsize) for fsize in filter_sizes]) self.dropout = nn.Dropout(dropout_prob) self.fc = nn.Linear(len(filter_sizes) * num_filters, output_dim) def forward(self, input_ids): embedded = self.embedding(input_ids).permute(0, 2, 1) # 转置以适应 Conv1D 输入 conved = [nn.functional.relu(conv(embedded)) for conv in self.convs] pooled = [nn.functional.max_pool1d(conv, conv.shape[2]).squeeze(2) for conv in conved] cat = torch.cat(pooled, dim=1) dropped = self.dropout(cat) logits = self.fc(dropped) return logits ``` #### 3. 训练过程 设置超参数并编写训练循环逻辑。 ```python def train_model(model, dataloader, loss_fn, optimizer, device): model.train() total_loss = 0 correct_preds = 0 total_samples = 0 for batch in dataloader: input_ids = batch['input_ids'].to(device) labels = batch['labels'].to(device) optimizer.zero_grad() outputs = model(input_ids) _, preds = torch.max(outputs, 1) loss = loss_fn(outputs, labels) loss.backward() optimizer.step() total_loss += loss.item() * input_ids.size(0) correct_preds += torch.sum(preds == labels.data) total_samples += input_ids.size(0) avg_loss = total_loss / total_samples accuracy = correct_preds.double() / total_samples return avg_loss, accuracy.item() ``` #### 4. 测试过程 测试阶段仅需计算损失与准确率即可。 ```python def evaluate_model(model, dataloader, loss_fn, device): model.eval() total_loss = 0 correct_preds = 0 total_samples = 0 with torch.no_grad(): for batch in dataloader: input_ids = batch['input_ids'].to(device) labels = batch['labels'].to(device) outputs = model(input_ids) _, preds = torch.max(outputs, 1) loss = loss_fn(outputs, labels) total_loss += loss.item() * input_ids.size(0) correct_preds += torch.sum(preds == labels.data) total_samples += input_ids.size(0) avg_loss = total_loss / total_samples accuracy = correct_preds.double() / total_samples return avg_loss, accuracy.item() ``` --- ### 参数配置与运行实例 假设我们有如下参数设定： - `vocab_size`: 字典大小。 - `embedding_dim`: 嵌入维度。 - `num_filters`: 卷积核数量。 - `filter_sizes`: 不同尺寸的一维卷积核。 - `output_dim`: 输出类别数（即标签种类）。 - `dropout_prob`: Dropout 层的概率。 ```python device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') model = TextCNN( vocab_size=len(word_to_idx), embedding_dim=100, num_filters=100, filter_sizes=[3, 4, 5], output_dim=10, dropout_prob=0.5 ).to(device) loss_fn = nn.CrossEntropyLoss() optimizer = torch.optim.Adam(model.parameters(), lr=0.001) train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True) test_loader = DataLoader(test_dataset, batch_size=64, shuffle=False) for epoch in range(10): # 设定迭代次数 train_loss, train_acc = train_model(model, train_loader, loss_fn, optimizer, device) test_loss, test_acc = evaluate_model(model, test_loader, loss_fn, device) print(f"Epoch {epoch+1}: Train Loss={train_loss:.4f}, Train Acc={train_acc:.4f}, Test Loss={test_loss:.4f}, Test Acc={test_acc:.4f}") ``` --- ### 总结 上述代码展示了如何使用 PyTorch 和 CNN 来实现 THUCNews 文本分类任务。其中重点在于利用卷积层提取局部特征的能力，从而提升分类效果[^3]。