基于TextRNN算法的文本二分类基础实现（Keras版）

数据准备：IMDb电影评论数据集+谷歌的Word2vec模型，（网上找不到的可以私聊我）

完整代码：

import os 
import time
import copy 
import gensim
import keras
import numpy as np
from sklearn.model_selection  import train_test_split
from keras.preprocessing.text import text_to_word_sequence,one_hot,Tokenizer
from keras.preprocessing.sequence import pad_sequences 
from keras.models import Sequential,Model 
from keras.layers import Input,Dense,GRU,LSTM,Activation,Dropout,Embedding
from keras.layers import Multiply,Concatenate,Dot 


#*************************1.读数据，获得X,Y*************************
datapath = r'./data/train'
pos_files = os.listdir(datapath + './pos')
neg_files = os.listdir(datapath + './neg')
pos_all = []
neg_all = []

for pf,nf in zip(pos_files,neg_files):
	with open(datapath + '/pos' + '/' + pf,encoding = 'utf-8') as f:
		s = f.read()
		pos_all.append(s)
	with open(datapath + '/neg' + '/' + nf,encoding = 'utf-8') as f:
		s = f.read()
		neg_all.append(s)

X_orig = np.array(pos_all + neg_all)
Y_orig = np.array([1 for _ in range(12500)] + [0 for _ in range(12500)]) #1表示pos,0表示neg

#*************************2.文本数值化*************************
#设置词典大小为20000，文本序列最大长度为200
#统计词频，过滤掉没有代表性的低频词
#生成词向量
#直接把词向量输入给模型，无法进行继续调优(fine-tune)
#应该加载词向量的同时，继续进行fine-tune 
vocab_size = 20000
maxlen     = 200
print("Start fitting the corpus......")
t = Tokenizer(vocab_size) 
ti = time.time()
t.fit_on_texts(X_orig) #输入x,训练语料库得到统计信息
to = time.time()
word_index = t.word_index #不受vocab_size的影响
print('all_vocab_size',len(word_index))
print('Fitting time:',(to - ti),'s')
print("Start vectorizing the sentences......")
v_X = t.texts_to_sequences(X_orig)
print("Start padding......")
pad_X = pad_sequences(v_X,maxlen = maxlen,padding = 'post')
print("Finished")

#*************************3.筛掉低频词，形成高频词字典*************************
x = list(t.word_counts.items())
s = sorted(x,key = lambda p:p[1],reverse = True)     #按词频排序
small_word_index = copy.deepcopy(word_index)         #防止原来的字典也被改变了
print("Removing less freq words from word_index dict......")
for item in s[20000:]:
	small_word_index.pop(item[0])
print("Finished")
print(len(small_word_index))
print(len(word_index))

#*************************4.词向量的导入*************************
#导入GoogleNews预训练好的300维word2vec词向量
model_file = './word2vec_model/GoogleNews-vectors-negative300.bin'
print("Loading word2vec model......")
w2v_model  = gensim.models.KeyedVectors.load_word2vec_format(model_file,binary = True,limit=10000) #不加limit会报Memory error
#把词向量，于word_index结合，构建一个embedding matrix 
"""
先随机初始化一个embedding matrix，这里需要注意的是，我们的词汇量vocab_size虽然是20000，
但是训练的时候还是会碰到不少词不在词汇表里，也在词向量也查不到，那这些词怎么处理呢？
我们就需要单独给这些未知词（UNK）一个index，在keras的文本预处理中，会默认保留index=0给这些未知词。
"""
embedding_matrix = np.random.uniform(size = (vocab_size + 1,300)) 
#(vocab_size + 1) * 300
#vocab_size + 1:+1是要留一个给index = 0
#300:300维的词向量
print("Transfering to the embedding matrix......")
#sorted_small_index = sorted(list(small_word_index.items()),key = lambda x:x[1])
for word,index in small_word_index.items():
	try:
		#词向量与单词索引之间的映射
		word_vector             = w2v_model[word]   
		embedding_matrix[index] = word_vector
	except:
		print("Word: [",word,"] not in word2vec_model ! Use random embedding instead.")

print("Finished")
print("Embedding matrix shape: \n",embedding_matrix.shape)

#*************************5.划分训练集和测试集*************************
np.random.seed = 1
random_indexs  = np.random.permutation(len(pad_X))
X = pad_X[random_indexs]
Y = Y_orig[random_indexs]
X_train,X_test,y_train,y_test = train_test_split(X,Y,test_size = 0.2)

#*************************6.模型搭建*************************
inputs = Input(shape = (maxlen,))
use_pretrained_w2v = True 
if use_pretrained_w2v:
	w2v = Embedding(vocab_size + 1,300,input_length = maxlen,weights = [embedding_matrix])(inputs)
else:
	w2v = Embedding(vocab_size + 1,300,input_length = maxlen)(inputs)

h = LSTM(128)(w2v)
y = Dense(1,activation = 'sigmoid')(h)
m = Model(input = inputs,output = y)
m.summary()  

m.compile(optimizer = 'adam',loss = 'binary_crossentropy',metrics = ['accuracy'])
m.fit(X_train,y_train,batch_size = 32,epochs = 3,validation_split = 0.15)

运行效果：

环境参数: Keras 2.1.2 + Tensorflow 1.4.1
参考教程： https://zhuanlan.zhihu.com/p/63852350