基于Keras搭建LSTM网络实现文本情感分类
一、语料概况1.1 数据统计1.1.1 查看样本均衡情况，对label进行统计1.1.2 计句子长度及长度出现的频数1.1.3 绘制句子长度累积分布函数(CDF) 二、LSTM模型构建与评估2.1 数据导入预处理2.1.1 获取标签和字符字典列表2.1.2 将文本和类别编码 2.2 构建LSTM模型2.3 划分数据集-LSTM模型训练-验证2.4 模型实际应用2.5 完整代码版本
在jupyter notbook中执行
import keras as ksimport tensorflow as tfprint(ks.__version__)print(tf.__version__)
2.3.1
2.0.0
一、语料概况
本次项目主要以电商网站下的用户评论数据作为实验数据集，数据集已经做好了标注。其中该数据集一共有4310条评论数据，文本的情感分为两类：“正面”和“反面”
其中evaluation为评论内容，label为情感倾向。
1.1 数据统计
先整体查看不同类别下的样本均衡情况，样本的不均衡会影响模型的分类效果
导入相应的库
import pandas as pdimport matplotlib.pyplot as pltfrom matplotlib import font_managerfrom itertools import accumulateplt.rcParams['font.sans-serif']=['SimHei'] #图中文字体设置为黑体
1.1.1 查看样本均衡情况，对label进行统计
df = pd.read_csv('./data_single.csv')print(df.groupby('label')['label'].count())
label正面 1908负面 2375Name: label, dtype: int64
整体来看样本还算均衡，正面1908条，负面2375条。可以达到需求了
深度学习中需要定义文本的最长长度，大于该长度就需要将文本切割，不够就填充0处理，所以还是需要整体统计出样本大致的文本长度分布。、
1.1.2 计句子长度及长度出现的频数
先计算句子的长度，在对句子的长度进行分组统计分析,看下不同长度下一共有几个样本
df[:30]df[:30].groupby('length').count()
可以看出长度15的有两个，length是句子长度，evaluation代表该长度下有几个样本，根据这两个值就可以画图了
df['length'] = df['evaluation'].apply(lambda x: len(x))len_df = df.groupby('length').count()sent_length = len_df.index.tolist()sent_freq = len_df['evaluation'].tolist()
获取数据中样本的所有长度sent_length，和该长度下sent_freq的样本数量绘图
# 绘制句子长度及出现频数统计图plt.figure(figsize=(10,12))plt.bar(sent_length, sent_freq,2)plt.title("句子长度及出现频数统计图", )plt.xlabel("句子长度")plt.ylabel("句子长度出现的频数")plt.savefig("./句子长度及出现频数统计图.png")plt.show
大部分长度在20左右。
1.1.3 绘制句子长度累积分布函数(CDF)
到句子累加到90%的时候，返回其对应长度值
### 绘制句子长度累积分布函数(CDF)sent_pentage_list = [(count/sum(sent_freq)) for count in accumulate(sent_freq)]plt.figure(figsize=(10,12))# 绘制CDFplt.plot(sent_length, sent_pentage_list)# 寻找分位点为quantile的句子长度quantile = 0.90for length, per in zip(sent_length, sent_pentage_list):if round(per, 2) == quantile:index = lengthbreakprint("\n分位点为%s的句子长度:%d." % (quantile, index))# 绘制句子长度累积分布函数图plt.plot(sent_length, sent_pentage_list,linewidth=4)plt.hlines(quantile, 0, index, colors="c", linestyles="dashed",linewidth=4)plt.vlines(index, 0, quantile, colors="c", linestyles="dashed",linewidth=4)plt.text(0, quantile, str(quantile),fontsize=10)plt.text(index, 0, str(index))plt.title("句子长度累积分布函数图",fontsize=20)plt.xlabel("句子长度",fontsize=15)plt.ylabel("句子长度累积频率",fontsize=15)plt.savefig("./句子长度累积分布函数图.png")plt.show
分位点为0.9的句子长度:172.
大多数样本的句子长度集中在1-200之间，句子长度累计频率取0.90分位点，则长度为172左右。
二、LSTM模型构建与评估
LSTM模型架构
导库
import pickleimport numpy as npimport pandas as pdfrom keras.utils import np_utils, plot_modelfrom keras.models import Sequentialfrom keras.preprocessing.sequence import pad_sequencesfrom keras.layers import LSTM, Dense, Embedding, Dropoutfrom sklearn.model_selection import train_test_splitfrom sklearn.metrics import accuracy_score
2.1 数据导入预处理
filepath = 'data_single.csv' #数据input_shape = 180 #设置长度model_save_path = 'corpus_model.h5' #模型保存位置df = pd.read_csv(filepath)df.head()
2.1.1 获取标签和字符字典列表
取出标签，和数据；再去重。
# 标签及词汇表labels, vocabulary = list(df['label'].unique()), list(df['evaluation'].unique())labels
[‘正面’, ‘负面’]
vocabulary
[‘用了一段时间，感觉还不错，可以’,
‘电视非常好，已经是家里的第二台了。第一天下单，第二天就到本地了，可是物流的人说车坏了，一直催，客服也帮着催，到第三天下午5点才送过来。父母年纪大了，买个大电视画面清晰，趁着耳朵还好使，享受几年。’,
‘电视比想象中的大好多，画面也很清晰，系统很智能，更多功能还在摸索中’,
‘不错’,
‘用了这么多天了，感觉还不错。夏普的牌子还是比较可靠。希望以后比较耐用，现在是考量质量的时候。’,
‘物流速度很快，非常棒，今天就看了电视，非常清晰，非常流畅，一次非常完美的购物体验’,
‘非常好，客服还特意打电话做回访’,
‘物流小哥不错，辛苦了，东西还没用’,
‘送货速度快，质量有保障，活动价格挺好的。希望用的久，不出问题。’,
‘非常漂亮，也非常清晰，反应速度也快。’,
‘很不错家里都喜欢。。。一次买了三台’,
‘送货非常快！质量非常好，这次购物非常愉快！！’,
‘58好大……都不错。看质量咯’,
‘物流很快，物有所值，值得信赖。依旧会关顾！谢谢商家！’,
‘这价钱超值，收到货马上装上看了一下。很清晰式样也蛮好！赞赞……’,
构建每个字符，去重后的集合。
# 构造字符级别的特征string = ''for word in vocabulary:#print(word)string += wordvocabulary = set(string)print(vocabulary)print(len(vocabulary))
2154
一共有2154个词，2个类别
对每个字符都进行编码
word_dictionary = {word: i+1 for i, word in enumerate(vocabulary)}word_dictionary
获取反转的字符编码
inverse_word_dictionary = {i+1: word for i, word in enumerate(vocabulary)}inverse_word_dictionary
设置词汇表大小,标签类别数量
vocab_size = len(word_dictionary.keys()) # 词汇表大小label_size = len(label_dictionary.keys()) # 标签类别数量
2.1.2 将文本和类别编码
将文本按照字典进行编码
x = [[word_dictionary[word] for word in sent] for sent in df['evaluation']]x
编码后，还需要切割填充，网络对于每一个样本数据的特征长度要求都是一致的
x = pad_sequences(maxlen=input_shape, sequences=x, padding='post', value=0)x
array([[1460, 325, 641, …, 0, 0, 0],
[1902, 2024, 905, …, 0, 0, 0],
[1902, 2024, , …, 0, 0, 0],
…,
[ 641, 703, 868, …, 0, 0, 0],
[ 633, 1902, 2024, …, 0, 0, 0],
[ 641, 1233, 1233, …, 0, 0, 0]])
类别需要编码成0,1的one_hot编码
y = [[label_dictionary[sent]] for sent in df['label']]y
[[0],
[0],
[0],
[0],
[0],
[0],
[0],
[0],
[0],
[0],
[0],
[0],
[0],
[0],
[0],
[0],
[0],
[0],
转为one_hot编码
y = [np_utils.to_categorical(label, num_classes=label_size) for label in y]y
array([[1., 0.]], dtype=float32),
array([[1., 0.]], dtype=float32),
array([[1., 0.]], dtype=float32),
array([[1., 0.]], dtype=float32),
array([[1., 0.]], dtype=float32),
array([[1., 0.]], dtype=float32),
array([[1., 0.]], dtype=float32),
array([[1., 0.]], dtype=float32),
array([[1., 0.]], dtype=float32),
array([[1., 0.]], dtype=float32),
array([[1., 0.]], dtype=float32),
array([[1., 0.]], dtype=float32),
array([[1., 0.]], dtype=float32),
转为标准化的输入
y = np.array([list(_[0]) for _ in y])y
array([[1., 0.],
[1., 0.],
[1., 0.],
…,
[0., 1.],
[0., 1.],
[0., 1.]], dtype=float32)
封装成函数,将字典保存下来
设置默认shape为20
# 导入数据# 文件的数据中，特征为evaluation, 类别为label.def load_data(filepath, input_shape=20):df = pd.read_csv(filepath)# 标签及词汇表labels, vocabulary = list(df['label'].unique()), list(df['evaluation'].unique())# 构造字符级别的特征string = ''for word in vocabulary:string += wordvocabulary = set(string)# 字典列表word_dictionary = {word: i+1 for i, word in enumerate(vocabulary)}with open('word_dict.pk', 'wb') as f:pickle.dump(word_dictionary, f)inverse_word_dictionary = {i+1: word for i, word in enumerate(vocabulary)}label_dictionary = {label: i for i, label in enumerate(labels)}with open('label_dict.pk', 'wb') as f:pickle.dump(label_dictionary, f)output_dictionary = {i: labels for i, labels in enumerate(labels)}vocab_size = len(word_dictionary.keys()) # 词汇表大小label_size = len(label_dictionary.keys()) # 标签类别数量# 序列填充，按input_shape填充，长度不足的按0补充x = [[word_dictionary[word] for word in sent] for sent in df['evaluation']]x = pad_sequences(maxlen=input_shape, sequences=x, padding='post', value=0)y = [[label_dictionary[sent]] for sent in df['label']]y = [np_utils.to_categorical(label, num_classes=label_size) for label in y]y = np.array([list(_[0]) for _ in y])return x, y, output_dictionary, vocab_size, label_size, inverse_word_dictionary
2.2 构建LSTM模型
可以看到中间的cell里面有四个黄色小框，你如果理解了那个代表的含义一切就明白了，每一个小黄框代表一个前馈网络层，对，就是经典的神经网络的结构，num_units就是这个层的隐藏神经元个数，就这么简单。其中1,2,4的激活函数是sigmoid，第三个的激活函数是tanh。
假设units = 64
根据上图，我们可以计算，假设a向量是128维向量，x向量是28维向量，那么二者concat以后就是156维向量，为了能相乘，那么Wf就应该是(64,156),同理其余三个框，也应该是同样的shape。于是，在第一层就有参数64x156x4 + 64x4个。
若是把cell外面的参数也算进去，那么假设有10个类，那么对于最终的shape为(64,1)的输出at,还要有一个shape为(10,64)的W跟一个shape为(10,1)的b。
# 模型输入参数，需要自己根据需要调整n_units = 100#LSTM神经元数量batch_size = 32#每次迭代数据量，一般为2的次方epochs = 5#训练批次output_dim = 20 #输出维度model = Sequential()model.add(Embedding(input_dim=vocab_size + 1, output_dim=output_dim,input_length=input_shape, mask_zero=True))model.add(LSTM(n_units, input_shape=(x.shape[0], x.shape[1])))model.add(Dropout(0.2))model.add(Dense(label_size, activation='softmax'))pile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])# plot_model(model, to_file='./model_lstm.png', show_shapes=True)model.summary()
模型架构：
Model: "sequential_8"_________________________________________________________________Layer (type) Output Shape Param # =================================================================embedding_7 (Embedding)(None, 180, 20) 43100_________________________________________________________________lstm_7 (LSTM)(None, 100)48400_________________________________________________________________dropout_1 (Dropout)(None, 100)0 _________________________________________________________________dense_7 (Dense) (None, 2) 202 =================================================================Total params: 91,702Trainable params: 91,702Non-trainable params: 0_________________________________________________________________
函数封装
# 创建深度学习模型， Embedding + LSTM + Softmax.def create_LSTM(n_units, input_shape, output_dim, filepath):x, y, output_dictionary, vocab_size, label_size, inverse_word_dictionary = load_data(filepath)model = Sequential()model.add(Embedding(input_dim=vocab_size + 1, output_dim=output_dim,input_length=input_shape, mask_zero=True))model.add(LSTM(n_units, input_shape=(x.shape[0], x.shape[1])))#model.add(Dropout(0.2))model.add(Dense(label_size, activation='softmax'))pile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])#plot_model(model, to_file='./model_lstm.png', show_shapes=True)model.summary()return model
2.3 划分数据集-LSTM模型训练-验证
用sk-learn划分数据集，9:1的比例
x, y, output_dictionary, vocab_size, label_size, inverse_word_dictionary = load_data(filepath, input_shape)train_x, test_x, train_y, test_y = train_test_split(x, y, test_size = 0.1, random_state = 42)# 模型输入参数，需要自己根据需要调整n_units = 100batch_size = 32epochs = 5output_dim = 20# 模型训练lstm_model = create_LSTM(n_units, input_shape, output_dim, filepath)lstm_model.fit(train_x, train_y, epochs=epochs, batch_size=batch_size, verbose=1)# 模型保存lstm_model.save(model_save_path)
Model: "sequential_9"_________________________________________________________________Layer (type) Output Shape Param # =================================================================embedding_8 (Embedding)(None, 180, 20) 43100_________________________________________________________________lstm_8 (LSTM)(None, 100)48400_________________________________________________________________dense_8 (Dense) (None, 2) 202 =================================================================Total params: 91,702Trainable params: 91,702Non-trainable params: 0_________________________________________________________________D:\Anaconda3\envs\tf3\lib\site-packages\tensorflow_core\python\framework\indexed_slices.py:424: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory."Converting sparse IndexedSlices to a dense Tensor of unknown shape. "Epoch 1/53854/3854 [==============================] - 15s 4ms/step - loss: 0.4679 - accuracy: 0.7706Epoch 2/53854/3854 [==============================] - 14s 4ms/step - loss: 0.2244 - accuracy: 0.9266Epoch 3/53854/3854 [==============================] - 14s 4ms/step - loss: 0.1719 - accuracy: 0.9463Epoch 4/53854/3854 [==============================] - 14s 4ms/step - loss: 0.1406 - accuracy: 0.9595Epoch 5/53854/3854 [==============================] - 15s 4ms/step - loss: 0.1274 - accuracy: 0.9606
测试数据
N = test_x.shape[0] # 测试的条数predict = []label = []for start, end in zip(range(0, N, 1), range(1, N+1, 1)):sentence = [inverse_word_dictionary[i] for i in test_x[start] if i != 0]y_predict = lstm_model.predict(test_x[start:end])label_predict = output_dictionary[np.argmax(y_predict[0])]label_true = output_dictionary[np.argmax(test_y[start:end])]print(''.join(sentence), label_true, label_predict) # 输出预测结果predict.append(label_predict)label.append(label_true)acc = accuracy_score(predict, label) # 预测准确率print('模型在测试集上的准确率为: %s.' % acc)
给家里老人买的，很不错哦，价格实惠 正面 正面
给父母买的，特意用了一段时间再来评价，电视非常好，没有坏点和损坏，界面也很简洁，便于操作，稍微不足就是开机会比普通电视慢一些，这应该是智能电视的通病吧，如果可以希望微鲸大大可以更新系统优化下开机时间~电视真的很棒，性价比爆棚，值得大家考虑购买。 客服很细心，快递小哥很耐心的等我通电验货，态度非常好。 负面 正面
长须鲸和海狮回答都很及时，虽然物流不够快但是服务不错电视不错，对比了乐视小米和微鲸论性价比还是微鲸好点 负面 负面
所以看不到4k效果，但是应该可以。 自带音响，中规中矩吧，好像没有别人说的好。而且，到现在没连接上我的漫步者，这个非常不满意，因为看到网上说好像普通3.5mm的连不上或者连上了声音小。希望厂家接下来开发的电视有改进。不知道我要不要换个音响。其他的用用再说。 放在地上的是跟我混了两年的tcl，天气受潮，修了一次，下岗了。 最后，我也觉得底座不算太稳，凑合着用。 负面 负面
电视机一般，低端机不要求那么高咯。 负面 负面
很好，两点下单上午就到了，服务很好。 正面 正面
帮朋友买的，好好好好好好好好 正面 正面
模型在测试集上的准确率为: 0.9254079254079254.
函数封装
# 模型训练def model_train(input_shape, filepath, model_save_path):# 将数据集分为训练集和测试集，占比为9:1# input_shape = 100x, y, output_dictionary, vocab_size, label_size, inverse_word_dictionary = load_data(filepath, input_shape)train_x, test_x, train_y, test_y = train_test_split(x, y, test_size = 0.1, random_state = 42)# 模型输入参数，需要自己根据需要调整n_units = 100batch_size = 32epochs = 5output_dim = 20# 模型训练lstm_model = create_LSTM(n_units, input_shape, output_dim, filepath)lstm_model.fit(train_x, train_y, epochs=epochs, batch_size=batch_size, verbose=1)# 模型保存lstm_model.save(model_save_path)N = test_x.shape[0] # 测试的条数predict = []label = []for start, end in zip(range(0, N, 1), range(1, N+1, 1)):sentence = [inverse_word_dictionary[i] for i in test_x[start] if i != 0]y_predict = lstm_model.predict(test_x[start:end])label_predict = output_dictionary[np.argmax(y_predict[0])]label_true = output_dictionary[np.argmax(test_y[start:end])]print(''.join(sentence), label_true, label_predict) # 输出预测结果predict.append(label_predict)label.append(label_true)acc = accuracy_score(predict, label) # 预测准确率print('模型在测试集上的准确率为: %s.' % acc)
2.4 模型实际应用
在模型实际应用的时候，需要导入对应的字符和标签字典，转为相应的编码。
# Import the necessary modulesimport pickleimport numpy as npfrom keras.models import load_modelfrom keras.preprocessing.sequence import pad_sequences# 导入字典with open('word_dict.pk', 'rb') as f:word_dictionary = pickle.load(f)with open('label_dict.pk', 'rb') as f:output_dictionary = pickle.load(f)try:# 数据预处理input_shape = 180sent = "很满意，电视非常好。护眼模式，很好，也很清晰。"x = [[word_dictionary[word] for word in sent]]print('--------x转为编码--------')x = pad_sequences(maxlen=input_shape, sequences=x, padding='post', value=0)print('--------x填充完成--------')# 载入模型model_save_path = './corpus_model.h5'lstm_model = load_model(model_save_path)# 模型预测y_predict = lstm_model.predict(x)label_dict = {v: k for k, v in output_dictionary.items()}print('输入语句: %s' % sent)print('情感预测结果: %s' % label_dict[np.argmax(y_predict)])except KeyError as err:print("您输入的句子有汉字不在词汇表中，请重新输入！")print("不在词汇表中的单词为：%s." % err)
--------x转为编码--------
--------x填充完成--------
输入语句: 很满意，电视非常好。护眼模式，很好，也很清晰。
情感预测结果: 正面
2.5 完整代码
统计出数据集中的情感分布以及评论句子长度分布
data_print.py
import pandas as pdimport matplotlib.pyplot as pltfrom matplotlib import font_managerfrom itertools import accumulate# 设置matplotlib绘图时的字体my_font = font_manager.FontProperties(fname='C:\Windows\Fonts\simfang.ttf')# 统计句子长度及出现次数的频数df = pd.read_csv('./data_single.csv')print(df.groupby('label')['label'].count())df['length'] = df['evaluation'].apply(lambda x: len(x))# print(df)len_df = df.groupby('length').count()sent_length = len_df.index.tolist()sent_freq = len_df['evaluation'].tolist()# 绘制句子长度及出现频数统计图plt.bar(sent_length, sent_freq)plt.title("句子长度及出现频数统计图", fontproperties=my_font)plt.xlabel("句子长度", fontproperties=my_font)plt.ylabel("句子长度出现的频数", fontproperties=my_font)plt.savefig("./句子长度及出现频数统计图.png")plt.close()# 绘制句子长度累计分布函数（CDF）sent_pentage_list = [(count / sum(sent_freq)) for count in accumulate(sent_freq)]# 绘制CDFplt.plot(sent_length, sent_pentage_list)# 寻找分位点为quantile的句子长度quantile = 0.91# print(list(sent_pentage_list))for length, per in zip(sent_length, sent_pentage_list):if round(per, 2) == quantile:index = lengthbreakprint('\n分位点为%s的句子长度：%d' % (quantile, index))# 绘制句子长度累积分布函数图plt.plot(sent_length, sent_pentage_list)plt.hlines(quantile, 0, index, colors="c", linestyles="dashed")plt.vlines(index, 0, quantile, colors="c", linestyles="dashed")plt.text(0, quantile, str(quantile))plt.text(index, 0, str(index))plt.title("句子长度累积分布函数图", fontproperties=my_font)plt.xlabel("句子长度", fontproperties=my_font)plt.ylabel("句子长度累积频率", fontproperties=my_font)plt.savefig("./句子长度累积分布函数图.png")plt.close()
import pickleimport numpy as npimport pandas as pdfrom keras.utils import np_utils, plot_modelfrom keras.models import Sequentialfrom keras.preprocessing.sequence import pad_sequencesfrom keras.layers import LSTM, Dense, Embedding, Dropoutfrom sklearn.model_selection import train_test_splitfrom sklearn.metrics import accuracy_score# 导入数据# 文件的数据中，特征为evaluation, 类别为label.def load_data(filepath, input_shape=20):df = pd.read_csv(filepath)# 标签及词汇表labels, vocabulary = list(df['label'].unique()), list(df['evaluation'].unique())# print(len(labels))# print(len(vocabulary))# 构造字符级别的特征string = ''for word in vocabulary:string += word# print(string)vocabulary = set(string)# print(vocabulary)# 字典列表word_dictionary = {word: i + 1 for i, word in enumerate(vocabulary)}with open('word_dict.pk', 'wb') as f:pickle.dump(word_dictionary, f)inverse_word_dictionary = {i + 1: word for i, word in enumerate(vocabulary)}label_dictionary = {label: i for i, label in enumerate(labels)}with open('label_dict.pk', 'wb') as f:pickle.dump(label_dictionary, f)output_dictionary = {i: labels for i, labels in enumerate(labels)}vocab_size = len(word_dictionary.keys()) # 词汇表大小label_size = len(label_dictionary.keys()) # 标签类别数量# print(vocab_size, labels)# 序列填充，按input_shape填充，长度不足的按0补充x = [[word_dictionary[word] for word in sent] for sent in df['evaluation']]x = pad_sequences(maxlen=input_shape, sequences=x, padding='post', value=0)y = [[label_dictionary[sent]] for sent in df['label']]y = [np_utils.to_categorical(label, num_classes=label_size) for label in y]y = np.array([list(_[0]) for _ in y])return x, y, output_dictionary, vocab_size, label_size, inverse_word_dictionary# 创建深度学习模型， Embedding + LSTM + Softmax.def create_LSTM(n_units, input_shape, output_dim, filepath):x, y, output_dictionary, vocab_size, label_size, inverse_word_dictionary = load_data(filepath)model = Sequential()model.add(Embedding(input_dim=vocab_size + 1, output_dim=output_dim,input_length=input_shape, mask_zero=True))model.add(LSTM(n_units, input_shape=(x.shape[0], x.shape[1])))model.add(Dropout(0.2))model.add(Dense(label_size, activation='softmax'))pile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])plot_model(model, to_file='./model_lstm.png', show_shapes=True)model.summary()return model# 模型训练def model_train(input_shape, filepath, model_save_path):# 将数据集分为训练集和测试集，占比为9:1# input_shape = 100x, y, output_dictionary, vocab_size, label_size, inverse_word_dictionary = load_data(filepath, input_shape)train_x, test_x, train_y, test_y = train_test_split(x, y, test_size=0.1, random_state=42)# 模型输入参数，需要自己根据需要调整n_units = 100batch_size = 32epochs = 5output_dim = 20# 模型训练lstm_model = create_LSTM(n_units, input_shape, output_dim, filepath)lstm_model.fit(train_x, train_y, epochs=epochs, batch_size=batch_size, verbose=1)# 模型保存lstm_model.save(model_save_path)N = test_x.shape[0] # 测试的条数predict = []label = []for start, end in zip(range(0, N, 1), range(1, N + 1, 1)):sentence = [inverse_word_dictionary[i] for i in test_x[start] if i != 0]y_predict = lstm_model.predict(test_x[start:end])label_predict = output_dictionary[np.argmax(y_predict[0])]label_true = output_dictionary[np.argmax(test_y[start:end])]print(''.join(sentence), label_true, label_predict) # 输出预测结果predict.append(label_predict)label.append(label_true)acc = accuracy_score(predict, label) # 预测准确率print('模型在测试集上的准确率为: %s.' % acc)if __name__ == '__main__':filepath = './data_single.csv'input_shape = 180# load_data(filepath, input_shape)model_save_path = './corpus_model.h5'model_train(input_shape, filepath, model_save_path)
模型预测代码
# Import the necessary modulesimport pickleimport numpy as npfrom keras.models import load_modelfrom keras.preprocessing.sequence import pad_sequences# 导入字典with open('word_dict.pk', 'rb') as f:word_dictionary = pickle.load(f)with open('label_dict.pk', 'rb') as f:output_dictionary = pickle.load(f)try:# 数据预处理input_shape = 180sent = "很满意，电视非常好。护眼模式，很好，也很清晰。"x = [[word_dictionary[word] for word in sent]]x = pad_sequences(maxlen=input_shape, sequences=x, padding='post', value=0)# 载入模型model_save_path = './corpus_model.h5'lstm_model = load_model(model_save_path)# 模型预测y_predict = lstm_model.predict(x)label_dict = {v: k for k, v in output_dictionary.items()}print('输入语句: %s' % sent)print('情感预测结果: %s' % label_dict[np.argmax(y_predict)])except KeyError as err:print("您输入的句子有汉字不在词汇表中，请重新输入！")print("不在词汇表中的单词为：%s." % err)
