【神经网络与深度学习】文本情感分类

数据准备

AclImdb -- v1 Dataset 是用于二进制情绪分类的大型电影评论数据集，其涵盖比基准数据集更多的数据，其中有 25,000 条电影评论用于训练，25,000 条用于测试，还有其他未经标记的数据可供使用。

数据预处理和数据装载

python 复制代码

import re

from torch.utils.data import DataLoader
from torch.utils.data import Dataset
import os

def tokenization(content):
    content = re.sub("<.*?>"," ",content)
    fileters = ['\t','\n','\x97','\x96','#','%','$','&',"\.","\?","!","\,"]
    content = re.sub("|".join(fileters)," ",content)
    tokens = [i.strip().lower() for i in content.split()]
    return tokens

def collate_fn(batch):
    """

    :param batch:( [tokens, labels], [tokens, labels])
    :return:
    """
    content, label  = list(zip(*batch))
    return content,label

class ImdbDataset(Dataset):
    def __init__(self, train=True):
        self.train_data_path = '..\\aclImdb\\train\\'
        self.test_data_path = '..\\aclImdb\\test\\'
        data_path = self.train_data_path if train else self.test_data_path
        #把所有文件名放入列表
        temp_data_path = [os.path.join(data_path,"pos"), os.path.join(data_path+"neg")]
        print(temp_data_path)
        self.total_file_path = [] #所有评论文件路径
        for path in temp_data_path:
            file_name_list = os.listdir(path)
            file_path_list = [os.path.join(path, i) for i in file_name_list if i.endswith(".txt")]
            self.total_file_path.extend(file_path_list)

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

    def __getitem__(self, index):
        file_path = self.total_file_path[index]
        # 获取label
        labelstr = file_path.split("\\")[-2]
        label = 0 if labelstr == "neg" else 1
        # 获取内容
        content = open(file_path).read()
        tokens = tokenization(content)


        return tokens, label

def get_data(train=True):
    imbd_dataset = ImdbDataset(train)
    data_loader = DataLoader(imbd_dataset, batch_size=2, shuffle=True,collate_fn=collate_fn)
    return data_loader

文本序列化

把文本里每个词语和其对应数字，使用字典保存即句子--->数字列表
思路：

句子进行分词（tokenization)
词语存入字典，统计出现次数，根据出现次数对齐进行过滤
把文本转数字序列
把数字序列转文本

遇到新出现的字符再词典里没有，可以用特殊字符替代

预保持每个batch里的序列大小一致，使用填充方法

python 复制代码

"""
构建词典 把句子转换成序列 再把序列转成句子
"""

class Word2Sequence:
    UNK_TAG = "UNK"
    PAD_TAG = "PAD"

    UNK =0
    PAD =1

    def __init__(self):
        self.dict = {
            self.UNK_TAG: self.UNK,
            self.PAD_TAG: self.PAD
        }
        self.count = {}

    def fit(self, sentence):

        # 把单个句子保存到dict
        for word in sentence:
            self.count[word] = self.count.get(word, 0)+1

    def build_vocab(self, min=5, max=None, max_features=None):
        """

        :param min:
        :param max:
        :param max_features: 一共保留多少个词语
        :return:
        """
        # 删除count中词频小于min的词语
        self.count = {word:value for word, value in self.count.items() if value>min}
        # 删除count中词频大于max的词语
        if max is not None:
            self.count = {word: value for word, value in self.count.items() if value < max}
        # 限制保留的词语数
        if max_features is not None:
            temp = sorted(self.cout.items(), key=lambda x:x[-1], reverse=True)[:max_features]
            self.count = dict(temp)

        # 把 词语 ------>数字
        for word in self.count:
            self.dict[word] = len(self.dict)

        # 得到一个反转的dict字典
        self.inverse_dict = dict(zip(self.dict.values(), self.dict.keys()))

    def transform(self, sentence, max_len=None):
        """
        把句子 转成 序列
        :param sentence:  [word1, word2, ..]
        :param max_len: 对句子进行填充或者裁剪
        :return:
        """
        if max_len is not None:
            if max_len > len(sentence):
                sentence = sentence + [self.PAD_TAG] * (max_len - len(sentence)) # 填充
            if max_len < len(sentence):
                sentence = sentence[:max_len] # 裁剪

        return [self.dict.get(word, self.UNK) for word in sentence]

    def inverse_transform(self, indices):
        # 把 序列 ------>句子
        return [self.inverse_dict.get(idx) for idx in indices]

if __name__ == '__main__':
    ws = Word2Sequence()
    ws.fit(["我","是","你","的","爸爸"])
    ws.fit(["我","是","我","的","人"])
    ws.build_vocab(min=0)

    print(ws.dict)
    re = ws.transform(["我","爱","人"],max_len=10)
    print(re)
    ret = ws.inverse_transform(re)
    print(ret)

模型构建（简单全连接）

注意 word_embedding的使用！

python 复制代码

"""
定义模型
"""
import torch
import torch.nn as nn
import torch.nn.functional as F
from lib import ws,max_len
from dataset import get_data
class MyModel(nn.Module):
    def __init__(self):
        super(MyModel, self).__init__()
        self.embedding = nn.Embedding(len(ws), 100)
        self.fc = nn.Linear(100*max_len, 2)



    def forward(self, input):
        """

        :param input: [batch_size, max_len]
        :return:
        """
        x = self.embedding(input) # [batch_size, max_len, 100]
        x = x.view([-1, 100*max_len])
        output = self.fc(x)
        return F.log_softmax(output,dim=-1)

model = MyModel()
optimizer = torch.optim.Adam(model.parameters(),lr=0.001)
def train(epoch):
    for idx,(input,target) in enumerate(get_data(train=True)):
        # 梯度清零
        optimizer.zero_grad()
        output= model(input)
        loss = F.nll_loss(output,target)
        loss.backward()
        optimizer.step()
        print(loss.item())

if __name__ == '__main__':
    for i in range(1):
        train(epoch=i)