Keras NLP——情感分析模型MLP

一.Python代码

#!/usr/bin/env python3
# encoding: utf-8
'''
@file: keras_emotional_analysis_mlp.py
@time: 2020/7/4 0004 12:06
@author: Jack
@contact: jack18588951684@163.com
'''
import string
import re
from os import listdir
from numpy import array
from nltk.corpus import stopwords
from keras.preprocessing.text import Tokenizer
from keras.utils.vis_utils import plot_model
from keras.models import Sequential
from keras.layers import Dense


## 加载文档
def load_doc(filename):
    file = open(filename, 'r')
    text = file.read()
    file.close()
    return text


## 清洗文档
def clean_doc(doc):
    tokens = doc.split()
    re_punc = re.compile('[%s]]' % re.escape(string.punctuation))
    tokens = [re_punc.sub('', w) for w in tokens]
    tokens = [w for w in tokens if w.isalpha()]
    stop_words = set(stopwords.words('english'))
    tokens = [w for w in tokens if not w in stop_words]
    tokens = [w for w in tokens if len(w) > 1]
    return tokens


## 编码文档
def doc_to_line(filename, vocab):
    doc = load_doc(filename)
    tokens = clean_doc(doc)
    tokens = [w for w in tokens if w in vocab]
    return ' '.join(tokens)


def process_docs(directory, vocab, is_train):
    lines = list()
    for filename in listdir(directory):
        if is_train and filename.startswith('cv9'):
            continue
        if not is_train and not filename.startswith('cv9'):
            continue
        path = directory + '/' + filename
        line = doc_to_line(path, vocab)
        lines.append(line)
    return lines


def load_clean_dataset(vocab, is_train):
    neg = process_docs('txt_sentoken/neg', vocab, is_train)
    pos = process_docs('txt_sentoken/pos', vocab, is_train)
    docs = neg + pos
    labels = array([0 for _ in range(len(neg))] + [1 for _ in range(len(pos))])
    return docs, labels


## 训练分词器
def create_tokenizer(lines):
    tokenizer = Tokenizer()
    tokenizer.fit_on_texts(lines)
    return tokenizer


## 定义模型
def define_model(n_words):
    model = Sequential()
    model.add(Dense(50, input_shape=(n_words,), activation='relu'))
    model.add(Dense(1, activation='sigmoid'))
    model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
    model.summary()
    # plot_model(model, to_file='model.png', show_shapes=True)
    return model


if __name__ == "__main__":
    ## 加载词典
    vocab_filename = 'vocab.txt'
    vocab = load_doc(vocab_filename)
    vocab = set(vocab.split())
    ## 加载所有评论文档
    train_docs, ytrain = load_clean_dataset(vocab, True)
    test_docs, ytest = load_clean_dataset(vocab, False)
    ## 创建分词器
    tokenizer = create_tokenizer(train_docs)
    ## 编码文档
    Xtrain = tokenizer.texts_to_matrix(train_docs, mode='freq')
    Xtest = tokenizer.texts_to_matrix(test_docs, mode='freq')
    ## 定义模型
    n_words = Xtest.shape[1]
    model = define_model(n_words)
    ## 训练模型
    model.fit(Xtrain, ytrain, epochs=10, verbose=2)
    ## 评估模型
    loss, acc = model.evaluate(Xtest, ytest, verbose=0)
    print('Test Accuracy:{}'.format(acc * 100))

二.代码说明

代码使用Keras中的Dense全连接层构建简单的多层感知器MLP将编码文档分类为正面或负面。模型的输入层的大小等于词汇表中的单词数即输入文档的长度，将其存储在一个名为n_words的新变量中。模型定义一个具有50个神经元和使用relu激活函数的隐藏层，输出层是使用sigmoid激活函数的单个神经元，用于预测最终评论的类别：0为负面评论，1为正面评论。模型训练过程中梯度下降使用Adam算法和二元交叉熵损失函数，在训练和评估模型时，跟踪模型准确性accuracy数据。

三.结果输出

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense (Dense)                (None, 50)                1181350   
_________________________________________________________________
dense_1 (Dense)              (None, 1)                 51        
=================================================================
Total params: 1,181,401
Trainable params: 1,181,401
Non-trainable params: 0
_________________________________________________________________
Epoch 1/10
57/57 - 0s - loss: 0.6912 - accuracy: 0.6300
Epoch 2/10
57/57 - 0s - loss: 0.6809 - accuracy: 0.8617
Epoch 3/10
57/57 - 0s - loss: 0.6613 - accuracy: 0.9172
Epoch 4/10
57/57 - 0s - loss: 0.6317 - accuracy: 0.9128
Epoch 5/10
57/57 - 0s - loss: 0.5948 - accuracy: 0.9383
Epoch 6/10
57/57 - 0s - loss: 0.5519 - accuracy: 0.9456
Epoch 7/10
57/57 - 0s - loss: 0.5067 - accuracy: 0.9456
Epoch 8/10
57/57 - 0s - loss: 0.4610 - accuracy: 0.9589
Epoch 9/10
57/57 - 0s - loss: 0.4187 - accuracy: 0.9561
Epoch 10/10
57/57 - 0s - loss: 0.3782 - accuracy: 0.9667
Test Accuracy:88.99999856948853