二十一、文本情感分类二

1.1 文本训练化概述

• 深度学习构建模型前需要将文本转化为向量表示（Word Embedding）。首先需要将文本转化为数字（文本序列化），在把数字转化为向量。
• 可以考虑把文本中的每个词语和其对应的数字，使用字典保存，同时把句子转化为数字的列表。

; 1.2 文本序列化的过程

• 实现文本序列化之前，应考虑一下几点：
• 如何使用字典把词语和数字进行对应；
• 不同的词语出现的次数不尽相同，是否需要对高频或者低频词语进行过滤
• 得到词典之后，如何把句子转化为数字序列，如何把数字序列转化为句子
• 不同句子长度不相同，每个batch的句子如何构造成相同的长度
• 对于新出现的词语在词典中没有出现怎么办（特殊字符代理）

2. 文本情感分类的模型构建

2.1 文本情感分类具体流程

1. 数据预处理：读取文本数据，并按照批量的方式加载数据，使用 DataLoader完成数据的读取，具体实现参考 data_prepare.py。
2. 文本序列化：将文本数据转化向量表示（Word Embedding），具体实现参考 save_ws.py。
3. 模型的构建和评估：由 model.py实现

2. 2 代码

• 步骤一：准备数据 data_prepare.py

import torch
from torch.utils.data import DataLoader, Dataset
import os
import re
import pickle

data_base_path = r"data\aclImdb"
ws = pickle.load(open("ws.pkl", "rb"))
max_len = 20

def tokenize(text):
    fileters = ['!', '"', '#', '$', '%', '&', '\(', '\)', '\*', '\+', ',', '-', '\.', '/', ':', ';', ', '=', '>',
                '\?', '@'
        , '\[', '\\', '\]', '^', '_', '`', '\{', '\|', '\}', '~', '\t', '\n', '\x97', '\x96', '"', '"', ]
    text = re.sub("", " ", text, flags=re.S)
    text = re.sub("|".join(fileters), " ", text, flags=re.S)
    return [i.strip() for i in text.split()]

class ImdbDataset(Dataset):
    def __init__(self, mode):

        super(ImdbDataset, self).__init__()
        if mode == "train":
            text_path = [os.path.join(data_base_path, i) for i in ["train/neg", "train/pos"]]
        else:
            text_path = [os.path.join(data_base_path, i) for i in ["test/neg", "test/pos"]]

        self.total_file_path_list = []
        for i in text_path:
            self.total_file_path_list.extend([os.path.join(i, j) for j in os.listdir(i)])

    def __getitem__(self, idx):
        cur_path = self.total_file_path_list[idx]

        cur_filename = os.path.basename(cur_path)
        label = int(cur_filename.split("_")[-1].split(".")[0]) - 1
        text = tokenize(open(cur_path, encoding="utf-8").read().strip())
        return label, text

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

def collate_fn(batch):

    label, content, = list(zip(*batch))
    content = [ws.transform(i, max_len=max_len) for i in content]
    content = torch.LongTensor(content)
    label = torch.LongTensor(label)
    return label, content

def get_dataloader(train_data=True):
    mode = ""
    if train_data:
        mode = "train"
    imdb_dataset = ImdbDataset(mode)
    dataloader = DataLoader(dataset=imdb_dataset, batch_size=10, shuffle=True, collate_fn=collate_fn)
    return dataloader

if __name__ == '__main__':

    text = "I cannot stay indifferent to Lars| van Trier's films. "
    s = tokenize(text)

    dataset = ImdbDataset(mode="train")

    dataloader = get_dataloader()

    for idx, (label, text) in enumerate(dataloader):
        print("idx：", idx)
        print("table:", label)
        print("text:", text)
        break

• 步骤二：文本序列化 save_ws.py

class ImdbDataset(Dataset):
    def __init__(self, mode):

        super(ImdbDataset, self).__init__()
        if mode == "train":
            text_path = [os.path.join(data_base_path, i) for i in ["train/neg", "train/pos"]]
        else:
            text_path = [os.path.join(data_base_path, i) for i in ["test/neg", "test/pos"]]

        self.total_file_path_list = []
        for i in text_path:
            self.total_file_path_list.extend([os.path.join(i, j) for j in os.listdir(i)])

    def __getitem__(self, idx):
        cur_path = self.total_file_path_list[idx]

        cur_filename = os.path.basename(cur_path)
        label = int(cur_filename.split("_")[-1].split(".")[0]) - 1
        text = tokenize(open(cur_path).read().strip())
        return label, text

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

• 步骤三：模型评估和预测 model.py

import pickle
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Adam
from unit21.data_prepare import get_dataloader, max_len

ws = pickle.load(open("ws.pkl", "rb"))

class IMDBModel(nn.Module):
    def __init__(self,max_len):
        super(IMDBModel,self).__init__()
        self.embedding = nn.Embedding(len(ws),300,padding_idx=ws.PAD)
        self.fc = nn.Linear(max_len*300,10)

    def forward(self, x):
        embed = self.embedding(x)
        embed = embed.view(x.size(0),-1)
        out = self.fc(embed)
        return F.log_softmax(out,dim=-1)

model = IMDBModel(max_len)
optimizer = Adam(model.parameters(), 0.001)

def train(epoch):
    train_dataloader = get_dataloader()
    for idx, (target, input) in enumerate(train_dataloader):
        optimizer.zero_grad()
        output = model(input)
        loss = F.nll_loss(output, target)
        loss.backward()
        optimizer.step()
        if idx % 10 == 0:
            print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
                epoch, idx * len(input), len(train_dataloader.dataset),
                       100. * idx / len(train_dataloader), loss.item()))
            torch.save(model.state_dict(), "imdb_net.pkl")

def test():
    test_loss = 0
    correct = 0
    train = False

    model.load_state_dict(torch.load("imdb_net.pkl"))

    model.eval()
    test_dataloader = get_dataloader()
    with torch.no_grad():
        for target, input in test_dataloader:
            output = model(input)
            test_loss += F.nll_loss(output, target, reduction="sum")
            pred = torch.max(output, dim=-1, keepdim=False)[-1]
            correct = pred.eq(target.data).sum()
        test_loss = test_loss / len(test_dataloader.dataset)
        print('\nTest set: Avg. loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)\n'.format(
            test_loss, correct, len(test_dataloader.dataset),
            100. * correct / len(test_dataloader.dataset)))

if __name__ == '__main__':
    epoch = 1
    train(epoch)

Original: https://blog.csdn.net/GUANGZHAN/article/details/122945121
Author: 智享AI
Title: 二十一、文本情感分类二