深度学习实现中文情感分析从获取数据集、预处理、构建模型、训练模型和测试模型的代码

获取数据集：

import pandas as pd

# 读取csv文件
df = pd.read_csv('data.csv', encoding='utf-8')
# 选择需要的列
df = df[['text', 'label']]
# 将标签转换为数字
df['label'] = df['label'].map({'positive': 1, 'negative': 0})

预处理：

import jieba

# 分词
def cut_text(text):
    return ' '.join(jieba.cut(text))

# 对文本进行分词
df['text'] = df['text'].apply(cut_text)

构建模型：

import torch
from torch import nn
from transformers import BertTokenizer, BertModel

class SentimentModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.bert = BertModel.from_pretrained('bert-base-chinese')
        self.fc = nn.Linear(768, 2)
        self.softmax = nn.Softmax(dim=1)

    def forward(self, x):
        _, pooled = self.bert(x)
        out = self.fc(pooled)
        out = self.softmax(out)
        return out

tokenizer = BertTokenizer.from_pretrained('bert-base-chinese')
model = SentimentModel()

训练模型：

from torch.utils.data import DataLoader, Dataset
import torch.optim as optim

class SentimentDataset(Dataset):
    def __init__(self, df, tokenizer):
        self.df = df
        self.tokenizer = tokenizer

    def __len__(self):
        return len(self.df)

    def __getitem__(self, idx):
        text = self.df.iloc[idx]['text']
        label = self.df.iloc[idx]['label']

        inputs = self.tokenizer.encode_plus(
            text,
            add_special_tokens=True,
            max_length=256,
            padding='max_length',
            truncation=True,
            return_attention_mask=True,
            return_tensors='pt'
        )

        return inputs['input_ids'], inputs['attention_mask'], torch.tensor(label)

def train(model, train_loader):
    optimizer = optim.Adam(model.parameters(), lr=1e-5)
    criterion = nn.CrossEntropyLoss()

    model.train()

    for epoch in range(10):
        for i, (input_ids, attention_mask, label) in enumerate(train_loader):
            optimizer.zero_grad()

            logits = model(input_ids=input_ids, attention_mask=attention_mask)
            loss = criterion(logits, label)

            loss.backward()
            optimizer.step()

            if i % 100 == 0:
                print('Epoch: {}/{} | Batch: {}/{} | Loss: {:.4f}'.format(
                    epoch+1, 10, i+1, len(train_loader), loss.item()))

train_dataset = SentimentDataset(df_train, tokenizer)
train_loader = DataLoader(train_dataset, batch_size=32, shuffle=True)

train(model, train_loader)

测试模型：

def predict(model, text):
    inputs = tokenizer.encode_plus(
        text,
        add_special_tokens=True,
        max_length=256,
        padding='max_length',
        truncation=True,
        return_attention_mask=True,
        return_tensors='pt'
    )

    logits = model(input_ids=inputs['input_ids'], attention_mask=inputs['attention_mask'])
    probs = nn.Softmax(dim=1)(logits)

    return probs.detach().numpy()[0]

text = '这部电影真的很好看，值得一看！'
probs = predict(model, text)
print('Positive probability:', probs[1])