Embedding + MLP：
Embedding：将经过独热编码过后的类别型特征向量化从而生成稠密特征
MLP：将稠密特征输入多层神经网络去拟合得到最优结果
数据集：
链接：https://pan.baidu.com/s/1e_98bghHp3X2hNoPoNn-sQ?pwd=ykaw
提取码：ykaw
导入数据+特征处理
- 对于类别型特征，如movieId userId movieGenre userGenre，先进行独热编码，再进行Embedding转换为稠密向量
- 对于数值型特征，直接输入MLP中
- tf.keras.layers.DenseFeatures 将特征列feature_columns转换为输入网络的张量dense tensor.
- tf.data.experimental.make_csv_dataset 将 CSV 文件读入数据集，其中数据集的每个元素都是一个（特征、标签）元组，对应于一批 CSV 行。 特征字典将特征列名称映射到包含相应特征数据的张量，标签是包含批次标签数据的张量。默认情况下，CSV 文件的第一行是列出列名的标题。 如果第一行不是标题，则设置 header=False 并使用 column_names 参数提供列名。默认情况下，数据集会无限重复，每次都会重新排列顺序。 可以通过设置 num_epochs 和 shuffle 参数来修改此行为。
参考：
https://andyguo.blog.csdn.net/article/details/121094641
import tensorflow as tf
TRAIN_DATA_URL = "file:///D:/SparrowRecSys-master/src/main/resources/webroot/sampledata/modelSamples.csv"
samples_file_path = tf.keras.utils.get_file("modelSamples.csv", TRAIN_DATA_URL)
TRAIN_DATA_URL1 = "file:///D:/SparrowRecSys-master/src/main/resources/webroot/sampledata/trainingSamples.csv"
samples_file_path1 = tf.keras.utils.get_file("trainingSamples.csv", TRAIN_DATA_URL1)
TRAIN_DATA_URL2 = "file:///D:/SparrowRecSys-master/src/main/resources/webroot/sampledata/testSamples.csv"
samples_file_path2 = tf.keras.utils.get_file("testSamples.csv", TRAIN_DATA_URL2)
#读入数据
def get_dataset(file_path):
dataset = tf.data.experimental.make_csv_dataset(
file_path,
batch_size=12,
label_name='label',
na_value="0",
num_epochs=1,
ignore_errors=True)
return dataset
#划分训练集和测试集
train_dataset = get_dataset(samples_file_path1)
test_dataset = get_dataset(samples_file_path2)
#载入类别型特征（类别型特征需要独热编码再Embedding化）
# movieId userId movieGenre userGenre
#风格类型
genre_vocab = ['Film-Noir', 'Action', 'Adventure', 'Horror', 'Romance', 'War', 'Comedy', 'Western', 'Documentary',
'Sci-Fi', 'Drama', 'Thriller',
'Crime', 'Fantasy', 'Animation', 'IMAX', 'Mystery', 'Children', 'Musical']
#用户喜欢的风格类型/电影风格 对应的风格类型词典
GENRE_FEATURES = {
'userGenre1': genre_vocab,
'userGenre2': genre_vocab,
'userGenre3': genre_vocab,
'userGenre4': genre_vocab,
'userGenre5': genre_vocab,
'movieGenre1': genre_vocab,
'movieGenre2': genre_vocab,
'movieGenre3': genre_vocab
}
# genre类别型特征（字符串型特征）转为one-hot特征，此处用到风格类型词表genre_vocab
categorical_columns = []
for feature, vocab in GENRE_FEATURES.items():
#独热编码
cat_col = tf.feature_column.categorical_column_with_vocabulary_list(
key=feature, vocabulary_list=vocab)
#向量化
emb_col = tf.feature_column.embedding_column(cat_col, 10)
categorical_columns.append(emb_col)
# 用户id和电影id（ID）转为one-hot特征，此处不用词表
#电影id独热编码
movie_col = tf.feature_column.categorical_column_with_identity(key='movieId', num_buckets=1001)
#电影id向量化
movie_emb_col = tf.feature_column.embedding_column(movie_col, 10)
categorical_columns.append(movie_emb_col)
#用户id独热编码
user_col = tf.feature_column.categorical_column_with_identity(key='userId', num_buckets=30001)
#用户id向量化
#为了将得到的one-hot转为稠密向量，所以要加一层embedding
user_emb_col = tf.feature_column.embedding_column(user_col, 10)
categorical_columns.append(user_emb_col)
#数值型特征 直接输入MLP内
#tf.feature_column.numeric_column 抽取数值型连续值
numerical_columns = [tf.feature_column.numeric_column('releaseYear'),
tf.feature_column.numeric_column('movieRatingCount'),
tf.feature_column.numeric_column('movieAvgRating'),
tf.feature_column.numeric_column('movieRatingStddev'),
tf.feature_column.numeric_column('userRatingCount'),
tf.feature_column.numeric_column('userAvgRating'),
tf.feature_column.numeric_column('userRatingStddev')]
搭建模型、模型评估
#搭建模型
model = tf.keras.Sequential()
#将数值型特征及类别型特征输入MLP
model.add(tf.keras.layers.DenseFeatures(numerical_columns + categorical_columns))
model.add(tf.keras.layers.Dense(128, activation='relu'))
model.add(tf.keras.layers.Dense(128, activation='relu'))
model.add(tf.keras.layers.Dense(1, activation='sigmoid'))
#编译模型
model.compile(
loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy', tf.keras.metrics.AUC(curve='ROC'),
tf.keras.metrics.AUC(curve='PR')])
model.fit(train_dataset, epochs=5)
test_loss, test_accuracy, test_roc_auc, test_pr_auc = model.evaluate(test_dataset)
print('\n\nTest Loss {}, Test Accuracy {}, Test ROC AUC {}, Test PR AUC {}'.format(test_loss, test_accuracy,test_roc_auc, test_pr_auc))
predictions = model.predict(test_dataset)
for prediction, goodRating in zip(predictions[:12], list(test_dataset)[0][1][:12]):
print("Predicted good rating: {:.2%}".format(prediction[0]),
" | Actual rating label: ",
("Good Rating" if bool(goodRating) else "Bad Rating"))
Epoch 1/5 7403/7403 [==============================] - 308s 41ms/step - loss: 3.5653 - accuracy: 0.5772 - auc_2: 0.5816 - auc_3: 0.6283 Epoch 2/5 7403/7403 [==============================] - 312s 42ms/step - loss: 0.6243 - accuracy: 0.6718 - auc_2: 0.7176 - auc_3: 0.7395 Epoch 3/5 7403/7403 [==============================] - 335s 45ms/step - loss: 0.5597 - accuracy: 0.7137 - auc_2: 0.7770 - auc_3: 0.7984 Epoch 4/5 7403/7403 [==============================] - 296s 40ms/step - loss: 0.5215 - accuracy: 0.7422 - auc_2: 0.8135 - auc_3: 0.8336 Epoch 5/5 7403/7403 [==============================] - 277s 37ms/step - loss: 0.4978 - accuracy: 0.7605 - auc_2: 0.8332 - auc_3: 0.8548 Predicted good rating: 79.00% | Actual rating label: Bad Rating Predicted good rating: 84.13% | Actual rating label: Good Rating Predicted good rating: 85.21% | Actual rating label: Bad Rating Predicted good rating: 57.63% | Actual rating label: Bad Rating Predicted good rating: 10.32% | Actual rating label: Bad Rating Predicted good rating: 45.51% | Actual rating label: Good Rating Predicted good rating: 28.17% | Actual rating label: Bad Rating Predicted good rating: 71.97% | Actual rating label: Bad Rating Predicted good rating: 77.62% | Actual rating label: Bad Rating Predicted good rating: 83.60% | Actual rating label: Bad Rating Predicted good rating: 70.58% | Actual rating label: Bad Rating Predicted good rating: 59.75% | Actual rating label: Bad Rating