Question

我是ML的新手，我正在根据我发现的教程，尝试使用Keras对IMDB数据集进行情感分析。下面的代码运行，并且测试数据的准确度约为90％。但是，当我尝试预测两个简单的句子（一个正数，一个负数）时，它的正值约为0.50，负数约为0.73，正数的值为0.71，负数的值为小于0.1，这是本教程中显示的结果。

有什么问题吗？

from keras.datasets import imdb
from keras.preprocessing import sequence
from keras.models import *
from keras.layers import *
import numpy as np

top_words = 5000  # 5000
# first tuple is data and sentiment lists,
# the second is testing data with sentiment
# https://keras.io/datasets/
(x_train, y_train), (x_test, y_test) = imdb.load_data(num_words=top_words)

# reverse lookup
word_to_id = imdb.get_word_index()
'''word_to_id = {k: (v + INDEX_FROM) for k, v in word_to_id.items()}'''
word_to_id["<PAD>"] = 0
word_to_id["<START>"] = 1
word_to_id["<UNK>"] = 2

# Truncate and pad the review sequences, to take only the first 500 words
max_review_length = 500
x_train = sequence.pad_sequences(x_train, maxlen=max_review_length)
x_test = sequence.pad_sequences(x_test, maxlen=max_review_length)

# Build the model

# embedding translates the words in a n dimensional
# space so "hi" becomes (0.2,0.1,0.5) in a 3 dimensional space
# it is the first layer of the network
embedding_vector_length = 32  # dimensions

# https://keras.io/getting-started/sequential-model-guide/
model = Sequential()  # sequential is a linear stack of layers

# layer of 500 x 32
model.add(
    Embedding(
        top_words,  # how many words to consider based on count
        embedding_vector_length,  # dimensions
        input_length=max_review_length))  # input vector
model.add(LSTM(100))  # the parameter are the memory units of the LSTM
# If you want you can replace LSTM by a flatten layer
# model.add(LSTM(100))
# model.add(Flatten())
model.add(Dense(1, activation='sigmoid'))  # output 0<y<1 for every x
model.compile(
    loss='binary_crossentropy',
    optimizer='adam',
    metrics=['accuracy'])
print(model.summary())


# Train the model
model.fit(
    x_train,
    y_train,
    validation_data=(x_test, y_test),
    epochs=1)  # original epochs = 3, batch-size = 64

# Evaluate the model
scores = model.evaluate(x_test, y_test, verbose=0)
print("Accuracy: %.2f%%" % (scores[1] * 100))

# predict sentiment from reviews
bad = "this movie was terrible and bad"
good = "i really liked the movie and had fun"
for review in [good, bad]:
    tmp = []
    for word in review.split(" "):
        tmp.append(word_to_id[word])
    tmp_padded = sequence.pad_sequences([tmp], maxlen=max_review_length)
    print("%s. Sentiment: %s" % (
        review, model.predict(np.array([tmp_padded[0]]))))
# i really liked the movie and had fun. Sentiment: 0.715537
# this movie was terrible and bad. Sentiment: 0.0353295

Answer 1

“有什么问题吗？”本身可能没有问题。我有一些想法，按照可能的影响顺序：

如果您的两个句子不能代表IMDB评论，那么您可以期望该模型的预测不准确且不规律。
您的模型只有一个时期，并且该模型可能没有足够的机会来学习从评论到情绪的可靠映射（假设在有数据的情况下可以进行这种映射）。
神经网络有一个随机元素，因此，您开发的模型可能无法与本教程中的模型进行相同的预测。
以“大约90％的准确度”，可以预期（取决于类别分布）十分之一的预测是不正确的。少数实例（在您的情况下为两个）通常不是评估模型性能的好方法。

当我运行您的代码时，我得到了大约80％的训练准确度和大约85％的测试准确度，并且“我真的很喜欢这部电影并且玩得开心。情感：[[0.75149596]]”和“这部电影糟糕和糟糕。情绪：[[0.93544275]]”。

Keras IMDB情绪分析

1 个答案: