Tensorflow学习笔记--使用keras完成文本情感分类问题

jilu

# -*- coding:utf-8 -*-
import numpy as np
import pandas as pd
import jieba
import csv

df = pd.read_csv('train_data.csv',encoding='utf-8')
df['label']=1
df.loc[df['satisfaction_id']==1,['label']]=1
df.loc[df['satisfaction_id']==3,['label']]=0
df['send_content_words'] = df['send_content'].apply(lambda s: list(jieba.cut(s))) #调用结巴分词

maxlen = 100 #截断词数
min_count = 5 #出现次数少于该值的词扔掉。这是最简单的降维方法

content = []
for i in df['send_content_words']:
	content.extend(i)
--diction
abc = pd.Series(content).value_counts()
abc = abc[abc >= min_count]

abc[:] = range(1, len(abc)+1)
abc[''] = 0

def doc2num(s, maxlen): 
    s = [i for i in s if i in abc.index]
    s = s[:maxlen] + ['']*max(0, maxlen-len(s))
    return list(abc[s])
df['doc2num'] = df['send_content_words'].apply(lambda s: doc2num(s, maxlen))

#手动打乱数据
idx = range(len(df))
np.random.shuffle(idx)
all_ = df.loc[idx]

#按keras的输入要求来生成数据
x = np.array(list(all_['doc2num']))
y = np.array(list(all_['label']))
y = y.reshape((-1,1)) #调整标签形状

from keras.models import Sequential
from keras.layers import Dense, Activation, Dropout, Embedding
from keras.layers import LSTM,Bidirectional

#建立模型
model = Sequential()
model.add(Embedding(len(abc), 64, input_length=maxlen))
model.add(Bidirectional(LSTM(128)))
model.add(Dropout(0.5))
model.add(Dense(1))
model.add(Activation('sigmoid'))
model.compile(loss='binary_crossentropy',
              optimizer='adam',
              metrics=['accuracy'])

batch_size = 128
train_num = 15000

model.fit(x, y, batch_size = batch_size, nb_epoch=30)

--test
df_test = pd.read_csv('test_data.csv',encoding='utf-8')
df_test['label']=1
df_test.loc[df['satisfaction_id']==1,['label']]=1
df_test.loc[df['satisfaction_id']==3,['label']]=0
df_test['send_content_words'] = df_test['send_content'].apply(lambda s: list(jieba.cut(s))) #调用结巴分词
df_test['doc2num'] = df_test['send_content_words'].apply(lambda s: doc2num(s, maxlen))
x_test = np.array(list(df_test['doc2num']))
y_test = np.array(list(df_test['label']))
y_test = y_test.reshape((-1,1)) #调整标签形状
model.evaluate(x_test,y_test)

Epoch 1/30
69999/69999 [==============================] - 75s - loss: 0.3004 - acc: 0.8809    
Epoch 2/30
69999/69999 [==============================] - 74s - loss: 0.2439 - acc: 0.9086    
Epoch 3/30
69999/69999 [==============================] - 74s - loss: 0.2239 - acc: 0.9175    
Epoch 4/30
69999/69999 [==============================] - 74s - loss: 0.2067 - acc: 0.9243    
Epoch 5/30
69999/69999 [==============================] - 74s - loss: 0.1888 - acc: 0.9326    
Epoch 6/30
69999/69999 [==============================] - 74s - loss: 0.1708 - acc: 0.9394    
Epoch 7/30
69999/69999 [==============================] - 74s - loss: 0.1556 - acc: 0.9445    
Epoch 8/30
69999/69999 [==============================] - 74s - loss: 0.1368 - acc: 0.9516    
Epoch 9/30
69999/69999 [==============================] - 74s - loss: 0.1248 - acc: 0.9562    
Epoch 10/30
69999/69999 [==============================] - 74s - loss: 0.1119 - acc: 0.9615    
Epoch 11/30
69999/69999 [==============================] - 75s - loss: 0.1007 - acc: 0.9655    
Epoch 12/30
69999/69999 [==============================] - 75s - loss: 0.0933 - acc: 0.9680    
Epoch 13/30
69999/69999 [==============================] - 75s - loss: 0.0849 - acc: 0.9711    
Epoch 14/30
69999/69999 [==============================] - 75s - loss: 0.0785 - acc: 0.9737    
Epoch 15/30
69999/69999 [==============================] - 75s - loss: 0.0651 - acc: 0.9786    
Epoch 16/30
69999/69999 [==============================] - 75s - loss: 0.0595 - acc: 0.9805    
Epoch 17/30
69999/69999 [==============================] - 75s - loss: 0.0555 - acc: 0.9821    
Epoch 18/30
69999/69999 [==============================] - 75s - loss: 0.0492 - acc: 0.9842    
Epoch 19/30
69999/69999 [==============================] - 76s - loss: 0.0499 - acc: 0.9839    
Epoch 20/30
69999/69999 [==============================] - 76s - loss: 0.0424 - acc: 0.9862    
Epoch 21/30
69999/69999 [==============================] - 75s - loss: 0.0402 - acc: 0.9871    
Epoch 22/30
69999/69999 [==============================] - 75s - loss: 0.0367 - acc: 0.9887    
Epoch 23/30
69999/69999 [==============================] - 75s - loss: 0.0353 - acc: 0.9885    
Epoch 24/30
69999/69999 [==============================] - 75s - loss: 0.0342 - acc: 0.9890    
Epoch 25/30
69999/69999 [==============================] - 74s - loss: 0.0273 - acc: 0.9913    
Epoch 26/30
69999/69999 [==============================] - 75s - loss: 0.0255 - acc: 0.9918    
Epoch 27/30
69999/69999 [==============================] - 75s - loss: 0.0289 - acc: 0.9905    
Epoch 28/30
69999/69999 [==============================] - 75s - loss: 0.0260 - acc: 0.9916    
Epoch 29/30
69999/69999 [==============================] - 75s - loss: 0.0244 - acc: 0.9920    
Epoch 30/30
69999/69999 [==============================] - 75s - loss: 0.0201 - acc: 0.9939 

df_test = pd.read_csv('test_data.csv',encoding='utf-8')
df_test['label']=1
df_test.loc[df['satisfaction_id']==1,['label']]=1
df_test.loc[df['satisfaction_id']==3,['label']]=0
df_test['send_content_words'] = df_test['send_content'].apply(lambda s: list(jieba.cut(s))) #调用结巴分词
df_test['doc2num'] = df_test['send_content_words'].apply(lambda s: doc2num(s, maxlen))
x_test = np.array(list(df_test['doc2num']))
y_test = np.array(list(df_test['label']))
y_test = y_test.reshape((-1,1)) #调整标签形状

model.evaluate(x_test,y_test)
[4.7231967521905895, 0.49716666666666665]

model.evaluate(x_test[:3000],y_test[:3000])

[1.0248313174247741, 0.84699999999999998]

#cnn
from keras.layers import Dense,Flatten,Reshape,AveragePooling2D,Convolution2D,Conv2D,MaxPooling2D
print('Build model...')
model = Sequential()
model.add(Embedding(len(abc), 64, input_length=maxlen))
model.add(Reshape(target_shape=(80,80,1),input_shape=(maxlen,64)))
model.add(Conv2D(32,(3,3),activation='relu',input_shape=(None,80,80,1),padding='same'))
model.add(Conv2D(32,(3,3),activation='relu'))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Dropout(0.25))

model.add(Conv2D(64,(3,3),activation='relu'))
model.add(Conv2D(64,(3,3),activation='relu'))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Dropout(0.25))

model.add(Flatten())
model.add(Dense(256,activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(1))
model.add(Activation('sigmoid'))
model.compile(loss='binary_crossentropy',
              optimizer='adam',
              metrics=['accuracy'])
batch_size = 128
train_num = 3500
model.fit(x,y, batch_size = batch_size, nb_epoch=30)