Load data

import pandas as pd
import numpy as np
# Import confusion_matrix and classification_report from the sklearn.metrics module
from sklearn.metrics import confusion_matrix, classification_report

df = pd.read_csv('./Restaurant_Reviews.tsv', sep='\t')
print(df)

Load model word2vec pretrained

import pandas as pd
from datetime import date, timedelta
import re
from nltk.tokenize import word_tokenize
from gensim import corpora, models
from gensim.models import KeyedVectors
from gensim.matutils import corpus2dense
import gensim

#Loading the word vectors from Google trained word2Vec model
GoogleModel = KeyedVectors.load_word2vec_format('./model/GoogleNews-vectors-negative300.bin', binary=True)

Test model

word = 'apple'
vector = GoogleModel[word]
print(vector.shape)

similar_words = GoogleModel.most_similar(word)
print(similar_words)

[('apples', 0.720359742641449), ('pear', 0.6450697183609009), ('fruit', 0.6410146355628967), ('berry', 0.6302294731140137),

Convert sang list token theo vocabulary

from keras_preprocessing.text import Tokenizer
from keras_preprocessing import sequence

tokenizer = Tokenizer()
tokenizer.fit_on_texts(df['Review'])
sequences = tokenizer.texts_to_sequences(df['Review'])

maxlen = 100
X = sequence.pad_sequences(sequences, maxlen=maxlen)
print('X:\n', X[:10])

Tạo ember maxtrix

import numpy as np

embedding_dim = 300
word_index = tokenizer.word_index
num_words = min(len(word_index) + 1, len(GoogleModel.index_to_key))
embedding_matrix = np.zeros((num_words, embedding_dim))

print('num_words:', num_words)
for word, i in word_index.items():
    if i >= num_words:
        continue
    if word in GoogleModel.index_to_key:
        embedding_matrix[i] = GoogleModel.word_vec(word)

Tạo model train

from keras.models import Sequential
from keras.layers import Embedding, Flatten, Dense, LSTM, Bidirectional

model = Sequential()
model.add(Embedding(num_words, embedding_dim, input_length=maxlen, weights=[embedding_matrix], trainable=False))
model.add(Bidirectional(LSTM(64, return_sequences=True, input_shape=(maxlen, ))))
model.add(Flatten())
model.add(Dense(1, activation='sigmoid'))

model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['acc'])
model.summary()

Model: "sequential_26"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 embedding_26 (Embedding)    (None, 100, 300)          621600    
                                                                 
 bidirectional_8 (Bidirectio  (None, 100, 128)         186880    
 nal)                                                            
                                                                 
 flatten_14 (Flatten)        (None, 12800)             0         
                                                                 
 dense_17 (Dense)            (None, 1)                 12801     
                                                                 
=================================================================
Total params: 821,281
Trainable params: 199,681
Non-trainable params: 621,600
_________________________________________________________________

Tạo data test

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, df['Liked'], test_size=0.3, random_state=101)

Train thôi.

model.fit(X_train, y_train, epochs=10, batch_size=32)

Test độ chính xác nồ

predictions = model.predict(X_test)
print('predict:', predictions[:3])
predictions = np.round(predictions)
print('predictions:', predictions.flatten())
print(confusion_matrix(y_test, predictions))
print(classification_report(y_test, predictions))

Done, code cứ thế mà run thôi , hy vọng sẽ giúp ích được cho mọi người . Thank for reading ,