Tolong ubah vektor representasi kata analisis sentimen CNN-LSTM berikut yang awalnya…
- User
Tolong ubah vektor representasi kata analisis sentimen CNN-LSTM berikut yang awalnya menggunakan Word2Vec menjadi menggunakan IndoBERT! "import numpy as np import pandas as pd import pickle from sklearn.model_selection import train_test_split from sklearn.preprocessing import LabelEncoder from sklearn.metrics import confusion_matrix, classification_report, accuracy_score from keras.preprocessing.text import Tokenizer from keras.utils import pad_sequences from keras.models import Sequential from keras.layers import Embedding, Conv1D, MaxPooling1D, LSTM, Dense, GlobalMaxPooling1D from keras.callbacks import EarlyStopping, ModelCheckpoint import gensim from gensim.models import Word2Vec from keras.layers import Dropout from keras.regularizers import l2 from keras.utils import to_categorical
Parameter untuk regularisasi L2
l2_coeff = 0.01
Membaca dataset
data_list = pickle.load(open('pre_processed_berita_121_joined_FIX_parpolheuristic_added.pkl', 'rb')) data = pd.DataFrame(data_list, columns=['judul', 'isi', 'pre_processed', 'Label', 'Partai_Politik_Heuristic']) data['Isi Berita'] = data['pre_processed']
Tokenisasi dan Padding
max_words = 10000 # mengurangi jumlah kata maksimum max_len = 500 # mengurangi panjang input tokenizer = Tokenizer(num_words=max_words) tokenizer.fit_on_texts(data['Isi Berita']) sequences = tokenizer.texts_to_sequences(data['Isi Berita']) X = pad_sequences(sequences, maxlen=max_len) y = to_categorical(data['Label'].astype('category').cat.codes)
Membagi data menjadi data latih dan data uji
X_train, X_test, y_train, y_test, idx_train, idx_test = train_test_split(X, y, data.index, test_size=0.25, random_state=42)
Membuka model word2vec
path = 'idwiki_word2vec_300.model' id_w2v = gensim.models.word2vec.Word2Vec.load(path) wv = id_w2v.wv
Membuat embeding matrix
embedding_dim = id_w2v.vector_size embedding_matrix = np.zeros((max_words, embedding_dim))
for word, i in tokenizer.word_index.items(): if i < max_words: try: embedding_vector = wv[word] embedding_matrix[i] = embedding_vector except KeyError: pass
Membangun model CNN-LSTM dengan pre-trained Word2Vec
num_filters = 64 lstm_units = 64
model = Sequential([ Embedding(input_dim=10000, output_dim=300, input_length=500, weights=[embedding_matrix], trainable=False), Conv1D(64, 5, activation='relu', kernel_regularizer=l2(0.01)), MaxPooling1D(pool_size=4), Dropout(0.25), LSTM(64, return_sequences=False, kernel_regularizer=l2(0.01), recurrent_regularizer=l2(0.01), dropout=0.2), Dense(32, activation='relu', kernel_regularizer=l2(0.01)), Dropout(0.5), Dense(3, activation='softmax') ])
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
Melatih model dengan early stopping dan model checkpoint
checkpoint = ModelCheckpoint('model_cnn_lstm.h5', save_best_only=True, monitor='val_accuracy', mode='max') early_stopping = EarlyStopping(monitor='val_loss', patience=10, verbose=1)
history = model.fit(X_train, y_train, validation_data=(X_test, y_test), epochs=50, batch_size=128)"
- Assistant
Berikut ini cara menggantikan Word2Vec dengan IndoBERT menggunakan Hugging Face Transformer:
Pertama, install library Hugging Face Transformer:
!pip install transformersLalu, ubah kode di atas dengan cara berikut (perhatikan bagian yang diubah):
import numpy as np import pandas as pd import pickle from sklearn.model_selection import train_test_split from sklearn.preprocessing import LabelEncoder from sklearn.metrics import confusion_matrix, classification_report, accuracy_score from keras.preprocessing.text import Tokenizer from keras.utils import pad_sequences from keras.models import Sequential from keras.layers import Embedding, Conv1D, MaxPooling1D, LSTM, Dense, GlobalMaxPooling1D from keras.callbacks import EarlyStopping, ModelCheckpoint from keras.layers import Dropout from keras.regularizers import l2 from keras.utils import to_categorical from transformers import BertTokenizer, TFBertModel # <-- Tambahkan ini # Parameter untuk regularisasi L2 l2_coeff = 0.01 # Membaca dataset data_list = pickle.load(open('pre_processed_berita_121_joined_FIX_parpolheuristic_added.pkl', 'rb')) data = pd.DataFrame(data_list, columns=['judul', 'isi', 'pre_processed', 'Label', 'Partai_Politik_Heuristic']) data['Isi Berita'] = data['pre_processed'] # Load pre-trained IndoBERT tokenizer dan model bert_model_name = 'indolem/indobert-base-uncased' tokenizer = BertTokenizer.from_pretrained(bert_model_name) bert_model = TFBertModel.from_pretrained(bert_model_name) # Tokenisasi menggunakan tokenizer IndoBERT input_encodings = tokenizer( data['Isi Berita'].to_list(), truncation=True, padding='max_length', max_length=128, return_tensors='tf' ) X = input_encodings['input_ids'] y = to_categorical(data['Label'].astype('category').cat.codes) # Membagi data menjadi data latih dan data uji X_train, X_test, y_train, y_test, idx_train, idx_test = train_test_split(X, y, data.index, test_size=0.25, random_state=42) # Membangun model CNN-LSTM dengan pre-trained IndoBERT model = Sequential([ bert_model.layers[0], # <-- Tambahkan ini (perhatikan ini adalah layer pertama dari "bert_model" di atas) Conv1D(64, 5, activation='relu', kernel_regularizer=l2(0.01)), MaxPooling1D(pool_size=4), Dropout(0.25), LSTM(64, return_sequences=False, kernel_regularizer=l2(0.01), recurrent_regularizer=l2(0.01), dropout=0.2), Dense(32, activation='relu', kernel_regularizer=l2(0.01)), Dropout(0.5), Dense(3, activation='softmax') ]) model.layers[0].trainable = False # <-- Jangan lupa menambahkan ini untuk membekukan layer bert model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy']) # Melatih model dengan early stopping dan model checkpoint checkpoint = ModelCheckpoint('model_cnn_lstm.h5', save_best_only=True, monitor='val_accuracy', mode='max') early_stopping = EarlyStopping(monitor='val_loss', patience=10, verbose=1) history = model.fit(X_train, y_train, validation_data=(X_test, y_test), epochs=50, batch_size=128)Perhatikan di atas kita menggunakan 128 sebagai jumlah token maksimum dari teks, Anda dapat menyesuaikannya sesuai kebutuhan Anda. Selain itu, CNN-LSTM akan menggunakan keluaran dari layer embedding BERT (yang telah dilatih) untuk analisis sentimen, daripada Word2Vec. Anda juga sebaiknya membekukan layer BERT selama pelatihan model CNN-LSTM untuk mengurangi TRAIN_TIME_MODEL & mengurangi konektifitas hingga 1. Anda bisa hapus line
model.layers[0].trainable = Falsejika Anda ingin melanjutkan fine-tune dengan CNN-LSTM Anda.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.