以下是基于 Transformer 的文本情感分析系统的完整代码，涵盖了数据预处理部分以及模型的选择与训练过程： ### 数据预处理 ```python from transformers import BertTokenizer from torch.utils.data import DataLoader, Dataset import pandas as pd class TextDataset(Dataset): def __init__(self, data_path, max_length=128): self.tokenizer = BertTokenizer.from_pretrained("bert-base-uncased") self.max_length = max_length # 加载并读取数据集 df = pd.read_csv(data_path) self.texts = df['text'].tolist() self.labels = df['label'].tolist() def __len__(self): return len(self.texts) def __getitem__(self, idx): text = str(self.texts[idx]) label = int(self.labels[idx]) encoding = self.tokenizer.encode_plus( text, add_special_tokens=True, max_length=self.max_length, padding='max_length', truncation=True, return_attention_mask=True, return_tensors='pt' ) return { 'input_ids': encoding['input_ids'].flatten(), 'attention_mask': encoding['attention_mask'].flatten(), 'labels': label } def create_data_loader(df, tokenizer, max_len, batch_size): ds = TextDataset( data_path=df, max_length=max_len ) return DataLoader(ds, batch_size=batch_size, num_workers=4) ``` 此段代码定义了一个 `TextDataset` 类用于加载和处理数据[^3]。通过 BERT 分词器对文本进行分割、填充和截断操作，确保所有输入序列长度一致。 ### 模型选择与训练 ```python from transformers import BertForSequenceClassification, AdamW import torch.nn.functional as F import torch.optim as optim from tqdm import tqdm device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') model = BertForSequenceClassification.from_pretrained("bert-base-uncased", num_labels=2).to(device) optimizer = AdamW(model.parameters(), lr=2e-5) def train_epoch(model, data_loader, optimizer, device): model.train() losses = [] correct_predictions = 0 for d in tqdm(data_loader): input_ids = d["input_ids"].to(device) attention_mask = d["attention_mask"].to(device) labels = d["labels"].to(device) outputs = model(input_ids=input_ids, attention_mask=attention_mask, labels=labels) loss = outputs.loss logits = outputs.logits _, preds = torch.max(logits, dim=1) correct_predictions += torch.sum(preds == labels) losses.append(loss.item()) loss.backward() optimizer.step() optimizer.zero_grad() accuracy = correct_predictions.double() / len(data_loader.dataset) return accuracy, sum(losses) / len(losses) EPOCHS = 3 for epoch in range(EPOCHS): print(f'Epoch {epoch + 1}/{EPOCHS}') accuracy, loss = train_epoch(model, dataloader_train, optimizer, device) print(f'Train Loss: {loss}, Accuracy: {accuracy}') ``` 该部分实现了基于 BERT 的分类模型，并使用交叉熵损失函数优化参数[^4]。整个流程包括前向传播计算损失值、反向传播更新权重等步骤。 ### 总结说明 以上代码展示了如何构建一个完整的基于 Transformer 的文本情感分析系统，其中涉及到了数据预处理方法如分词、填充及标签编码等内容[^1]；同时也包含了具体实现细节比如选用何种框架完成建模任务等问题解答[^2]。