Keras NLP——預測新評論的情感

一.Python代碼

#!/usr/bin/env python3
# encoding: utf-8
'''
@file: Keras_predict_sentiment.py
@time: 2020/7/5 0005 11:58
@author: Jack
@contact: [email protected]
'''

import string
import re
from os import listdir
from numpy import array
from nltk.corpus import stopwords
from keras.preprocessing.text import Tokenizer
from keras.utils.vis_utils import plot_model
from keras.models import Sequential
from keras.layers import Dense


def load_doc(filename):
    file = open(filename, 'r')
    text = file.read()
    file.close()
    return text


def clean_doc(doc):
    tokens = doc.split()
    re_punc = re.compile('[%s]' % re.escape(string.punctuation))
    tokens = [re_punc.sub('', w) for w in tokens]
    tokens = [w for w in tokens if w.isalpha()]
    stop_words = set(stopwords.words('english'))
    tokens = [w for w in tokens if not w in stop_words]
    tokens = [w for w in tokens if len(w) > 1]
    return tokens


def doc_to_line(filename, vocab):
    doc = load_doc(filename)
    tokens = clean_doc(doc)
    tokens = [w for w in tokens if w in vocab]
    return ' '.join(tokens)


def precess_docs(directory, vocab):
    lines = list()
    for filename in listdir(directory):
        path = directory + '/' + filename
        line = doc_to_line(path, vocab)
        lines.append(line)
    return lines


def load_clean_dataset(vocab):
    neg = precess_docs('txt_sentoken/neg', vocab)
    pos = precess_docs('txt_sentoken/pos', vocab)
    docs = neg + pos
    labels = array([0 for _ in range(len(neg))] + [1 for _ in range(len(pos))])
    return docs, labels


def create_tokenizer(lines):
    tokenizer = Tokenizer()
    tokenizer.fit_on_texts(lines)
    return tokenizer


def define_model(n_words):
    model = Sequential()
    model.add(Dense(50, input_shape=(n_words,), activation='relu'))
    model.add(Dense(1, activation='sigmoid'))
    model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
    model.summary()
    # plot_model(model, to_file='model.png', show_shapes=True)
    return model


def predict_sentiment(review, vocab, tokenizer, model):
    tokens = clean_doc(review)
    tokens = [w for w in tokens if w in vocab]
    line = ' '.join(tokens)
    encoded = tokenizer.texts_to_matrix([line], mode='binary')
    yhat = model.predict(encoded, verbose=0)
    percent_pos = yhat[0, 0]
    if round(percent_pos) == 0:
        return (1 - percent_pos), 'NEGATIVE'
    return percent_pos, 'POSITIVE'


if __name__ == "__main__":
    vocab_filename = 'vocab.txt'
    vocab = load_doc(vocab_filename)
    vocab = set(vocab.split())
    train_docs, ytrain = load_clean_dataset(vocab)
    test_docs, ytest = load_clean_dataset(vocab)
    tokenizer = create_tokenizer(train_docs)
    Xtrain = tokenizer.texts_to_matrix(train_docs, mode='binary')
    Xtest = tokenizer.texts_to_matrix(test_docs, mode='binary')
    n_words = Xtrain.shape[1]
    model = define_model(n_words)
    model.fit(Xtrain, ytrain, epochs=10, verbose=2)
    text = 'Best movie ever! It was great!.'
    percent, sentiment = predict_sentiment(text, vocab, tokenizer, model)
    print("Review:{}\tSentiment:{}({})".format(text, sentiment, percent * 100))

    text = 'This is a bad movie.'
    percent, sentiment = predict_sentiment(text, vocab, tokenizer, model)
    print("Review:{}\tSentiment:{}({})".format(text, sentiment, percent * 100))

二.代碼說明

最後使用所有可用數據（訓練集+測試集）訓練開發最終模型並使用其來預測新的電影評論的類別。預測新評論的情感分類同樣需要遵循相同的測試數據準備步驟，即加載文檔、清理文檔、編碼文檔，然後進行預測。代碼中體現爲函數predict_sentiment()，函數的參數爲評論文本、詞彙表、分詞器tokenizer和模型，返回爲預測的情緒和該分類在兩個分類中的百分比，通過調用predict()來直接使用擬合模型預測分類值，對於負面評論返回0，正面評論返回1。

三.結果輸出

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense (Dense)                (None, 50)                1288950   
_________________________________________________________________
dense_1 (Dense)              (None, 1)                 51        
=================================================================
Total params: 1,289,001
Trainable params: 1,289,001
Non-trainable params: 0
_________________________________________________________________
Epoch 1/10
63/63 - 0s - loss: 0.4540 - accuracy: 0.7920
Epoch 2/10
63/63 - 0s - loss: 0.0531 - accuracy: 0.9965
Epoch 3/10
63/63 - 0s - loss: 0.0151 - accuracy: 1.0000
Epoch 4/10
63/63 - 0s - loss: 0.0069 - accuracy: 1.0000
Epoch 5/10
63/63 - 0s - loss: 0.0037 - accuracy: 1.0000
Epoch 6/10
63/63 - 1s - loss: 0.0020 - accuracy: 1.0000
Epoch 7/10
63/63 - 1s - loss: 0.0012 - accuracy: 1.0000
Epoch 8/10
63/63 - 0s - loss: 7.2241e-04 - accuracy: 1.0000
Epoch 9/10
63/63 - 0s - loss: 4.8940e-04 - accuracy: 1.0000
Epoch 10/10
63/63 - 0s - loss: 3.5379e-04 - accuracy: 1.0000
Review:Best movie ever! It was great!.	Sentiment:POSITIVE(51.90609693527222)
Review:This is a bad movie.	Sentiment:NEGATIVE(67.59185492992401)