實戰電影評論情感分析

        情感分析是一段文字表達的情緒狀態。其中，一段文字可以使一個句子、一個段落或者一個文件。主要涉及兩個問題：文字表達和文字分類。在深度學習出現之前，主流的表示方法有BOW(詞袋模型)和topic model(主題模型)，分類模型主要有SVM和LR。

         載入資料：IMDB情感分析資料集，訓練集和測試集分別包含了25000條已標註的電影評論，滿分了10分，小於等於4為負面評論。

# -*- coding: utf-8 -*-

import numpy as np 
# 載入已訓練好的詞典向量模型，包含400000的文字向量，每行有50維的資料
words_list = np.load('wordsList.npy')
print('載入word列表')
words_list = words_list.tolist()   # 轉化為list
words_list = [word.decode('UTF-8') for word in words_list]
word_vectors = np.load('wordVectors.npy')
print('載入文字向量')

print(len(words_list))
print(word_vectors.shape)

Home_index = words_list.index("home")
print(word_vectors[Home_index])

# 載入電影資料
import os
from os.path import isfile, join
pos_files = ['pos/' + f for f in os.listdir('pos/') if isfile(join('pos/', f))]
neg_files = ['neg/' + f for f in os.listdir('neg/') if isfile(join('neg/', f))]
num_words = []
for pf in pos_files:
	with open(pf, "r", encoding='utf-8') as f:
		line = f.readline()
		counter = len(line.split())
		num_words.append(counter)
print('正面評價完結')

for pf in neg_files:
	with open(pf, "r", encoding='utf-8') as f:
		line = f.readline()
		counter = len(line.split())
		num_words.append(counter)
print('負面評價完結')

num_files = len(num_words)
print('檔案總數', num_files)
print('所有的詞的數量', sum(num_words))
print('平均檔案詞的長度', sum(num_words)/len(num_words))

'''
# 視覺化
import matplotlib
import matplotlib.pyplot as plt
matplotlib.use('qt4agg')
# 指定預設字型
matplotlib.rcParams['font.sans-serif'] = ['SimHei']
matplotlib.rcParams['font.family'] = 'sans-serif'
#%matplotlib inline
plt.hist(num_words, 50, facecolor='g')
plt.xlabel('文字長度')
plt.ylabel('頻次')
plt.axis([0, 1200, 0, 8000])
plt.show()
'''

# 大部分文字都在230之內
max_seg_len = 300

# 將文字生成一個索引矩陣，得到一個25000x300矩陣
import re
strip_special_chars = re.compile("[^A-Za-z0-9 ]+")

def cleanSentence(string):
	string = string.lower().replace("<br />", " ")
	return re.sub(strip_special_chars, "", string.lower())
print('儲存idxMatrix...')
max_seg_num = 300
ids = np.zeros((num_files, max_seg_num), dtype="int32")
file_count = 0
'''
for pf in pos_files:
	with open(pf, "r", encoding="utf-8") as f:
		indexCounter = 0
		line = f.readline()
		cleanedLine = cleanSentence(line)
		split = cleanedLine.split()
		for word in split:
			try:
				ids[file_count][indexCounter] = words_list.index(word)
			except ValueError:
				ids[file_count][indexCounter] = 399999 # 未知的詞
			indexCounter = indexCounter + 1
			if indexCounter >= max_seg_num:
				break
		file_count = file_count + 1
		print(file_count)
print('儲存完成1')
for nf in neg_files:
	with open(nf, "r", encoding="utf-8") as f:
		indexCounter = 0
		line = f.readline()
		cleanedLine = cleanSentence(line)
		split = cleanedLine.split()
		for word in split:
			try:
				ids[file_count][indexCounter] = words_list.index(word)
			except ValueError:
				ids[file_count][indexCounter] = 399999 # 未知的詞
			indexCounter = indexCounter + 1
			if indexCounter >= max_seg_num:
				break
		file_count = file_count + 1

# 儲存到檔案
np.save('idxMatrix', ids)
print('儲存完成2')
'''

# 模型設定
batch_size = 24
lstm_units = 64
num_labels = 2
iterations = 200000
max_seg_num = 250
ids = np.load('idsMatrix.npy')

# 返回一個數據集的迭代器， 返回一批訓練集合
from random import randint
def get_train_batch():
	labels = []
	arr = np.zeros([batch_size, max_seg_num])
	for i in range(batch_size):
		if (i % 2 == 0):
			num = randint(1, 11499)
			labels.append([1, 0])
		else:
			num = randint(13499, 24999)
			labels.append([0, 1])
		arr[i] = ids[num-1: num]
	return arr, labels

def get_test_batch():
	labels = []
	arr = np.zeros([batch_size, max_seg_num])
	for i in range(batch_size):
		num = randint(11499, 13499)
		if (num <= 12499):
			labels.append([1, 0])
		else:
			labels.append([0, 1])
		arr[i] = ids[num-1:num]
	return arr, labels

num_dimensions = 300  # Dimensions for each word vector

import tensorflow as tf
tf.reset_default_graph()
labels = tf.placeholder(tf.float32, [batch_size, num_labels])
input_data = tf.placeholder(tf.int32, [batch_size, max_seg_num])

data = tf.Variable(tf.zeros([batch_size, max_seg_num, num_dimensions]), dtype=tf.float32)
data = tf.nn.embedding_lookup(word_vectors, input_data)

# 配置LSTM網路
lstmCell = tf.contrib.rnn.BasicLSTMCell(lstm_units)
lstmCell = tf.contrib.rnn.DropoutWrapper(cell=lstmCell, output_keep_prob=0.75) # 避免一些過擬合
value, _ = tf.nn.dynamic_rnn(lstmCell, data, dtype=tf.float32)

# 第一個輸出可以被認為是最後的隱藏狀態，該向量將重新確定維度，然後乘以一個權重加上偏置，獲得最終的label
weight = tf.Variable(tf.truncated_normal([lstm_units, num_labels]))
bias = tf.Variable(tf.constant(0.1, shape=[num_labels]))
value = tf.transpose(value, [1, 0, 2])
last = tf.gather(value, int(value.get_shape()[0]) - 1)
prediction = (tf.matmul(last, weight) + bias)

# 預測函式以及正確率評估引數
correct_pred = tf.equal(tf.argmax(prediction, 1), tf.argmax(labels, 1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

# 將標準的交叉熵損失函式定義為損失值，選擇Adam作為優化函式
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=prediction, labels=labels))
optimizer = tf.train.AdamOptimizer().minimize(loss)

#sess = tf.InteractiveSession(config=tf.ConfigProto(allow_soft_placement, log_device_placement))
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True, log_device_placement=False))

#saver = tf.train.Saver()
#saver.restore(sess, tf.train.latest_checkpoint('models'))

iterations = 10
for i in range(iterations):
  next_batch, next_batch_labels = get_test_batch()
  print("正確率:", (sess.run(
      accuracy, {input_data: next_batch, labels: next_batch_labels})) * 100)

'''
# 使用tensorboard視覺化損失值和正確值
import datetime
sess = tf.InteractiveSession()
#tf.device("/cpu:0")
saver = tf.train.Saver()
sess.run(tf.global_variables_initializer())

tf.summary.scalar('Loss', loss)
tf.summary.scalar('Accuracy', accuracy)
merged = tf.summary.merge_all()
logdir = "tensorboard/" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S") + "/"
writer = tf.summary.FileWriter(logdir, sess.graph)

for i in range(iterations):
	# 下個批次的資料
	nextBatch, nextBatchLabels = get_train_batch();
	sess.run(optimizer, {input_data: nextBatch, labels: nextBatchLabels})
	# 每50次寫入一次leadboard
	if (i % 50 == 0):
		summary = sess.run(merged, {input_data: nextBatch, labels: nextBatchLabels})
		writer.add_summary(summary, i)
	# 每10,000次儲存一個模型
	if (i % 10000 == 0 and i != 0):
		save_path = saver.save(sess, "models/pretrained_lstm.ckpt", global_step=i)
		print("saved to %s" % save_path)

writer.close()
'''