# -*- coding: utf-8 -*-
import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Embedding, LSTM, Dense, Dropout, Bidirectional
from tensorflow.keras.optimizers import Adam
import jieba
def cut_text(text):
seg_list = jieba.cut(text)
return ' '.join(seg_list)
df = pd.read_csv('weibo_senti_100k.csv', nrows=50000)
# 分词
word_cut = df['review'].map(cut_text)
# # 分割数据集
x_train, x_test, y_train, y_test = train_test_split(word_cut, df['label'], test_size=0.2, random_state=42)
# 文本数据转换为序列
tokenizer = Tokenizer()
tokenizer.fit_on_texts(word_cut)
# print(tokenizer.word_index)
X_train_seq = tokenizer.texts_to_sequences(x_train)
X_test_seq = tokenizer.texts_to_sequences(x_test)
# 填充序列
max_len = max(len(x) for x in X_train_seq)
print(max_len)
X_train_pad = pad_sequences(X_train_seq, maxlen=120, padding='post')
X_test_pad = pad_sequences(X_test_seq, maxlen=120, padding='post')
# 将标签转换为numpy数组
# y_train = np.array(y_train)
# y_test = np.array(y_test)
#构建更复杂的模型
vocab_size = len(tokenizer.word_index) + 1 # 词汇表大小
embedding_dim = 100 # 词嵌入维度
lstm_units = 128 # LSTM单元数
model = Sequential()
model.add(Embedding(input_dim=vocab_size, output_dim=embedding_dim, input_length=max_len))
# model.add(Bidirectional(LSTM(lstm_units, return_sequences=True)))
model.add(Dropout(0.5))
model.add(Bidirectional(LSTM(lstm_units)))
model.add(Dense(1, activation='sigmoid'))
# # 编译模型
optimizer = Adam(learning_rate=1e-3)
model.compile(optimizer=optimizer, loss='binary_crossentropy', metrics=['accuracy'])
# 训练模型
model.fit(X_train_pad, y_train, epochs=10, batch_size=200, validation_data=(X_test_pad, y_test))
# 评估模型
loss, accuracy = model.evaluate(X_test_pad, y_test)
print(f'测试集上的准确率: {accuracy:.2f}')
model.save('my_model.keras')
# 测试模型
new_sentences = [
'这手机用起来真是太流畅了，推荐！',
'看了这部电影，感觉浪费了两个小时。',
'新出的餐厅环境很好，菜品也很棒，值得一试！',
'快递太慢了，等了一个星期还没到。',
'今天的天气真好，心情也变得很美丽。',
'这次更新的功能非常实用，点赞！',
'真搞不懂这些人怎么想的，这样也能接受？',
'旅游景点人太多了，根本没法好好玩。',
'听说这个品牌的化妆品很不错，打算试试。',
'真是失望，服务态度这么差，还怎么做生意。'
]
new_sentences = pd.Series(new_sentences)
new_sentences_cut = new_sentences.map(cut_text)
new_sequences2 = tokenizer.texts_to_sequences(new_sentences_cut)
new_padded = pad_sequences(new_sequences2, maxlen=120, padding='post')
predictions = model.predict(new_padded)
predictions = model.predict(new_padded)
predictions = [1 if p > 0.5 else 0 for p in predictions]
for sentence, sentiment in zip(new_sentences, predictions):
print(f'句子: {sentence} -> 情感: {"积极" if sentiment == 1 else "消极"}')
