Hi everybody i have developed two RNN models for a chatbot.Let's say that user says:"Tell me how the weather will be tomorrow in Paris". The first model will be able to recognize the user's intent WEATHER_INFO , while the second one will be able to extract meaningful information from the phrase like LOC:Paris and DATE:tomorrow Of course there are many other intent categories like for example MUSIC_PLAY and so on. Since the two model are not linked in any way for the same phrase we could have for example the results intent:WEATHER_INFO and TITLE:Paris , where the slot TITLE is instead linked to the MUSIC_PLAY intention Many researchers tried to improve performance creating a joint model where the two models use the other one in order to avoid this kind of error here my code for the slot filling nn
from keras.models import Model, Input
from keras.layers import LSTM, Embedding, Dense, TimeDistributed, Dropout, Bidirectional
import keras as k
from keras_contrib.layers import CRF
# input = Input(shape=(140,))
# input = Input(shape=(len(X_train),max_len))
input = Input(shape=(max_len,))
# word_embedding_size = 150
word_embedding_size = 150
n_words = len(token_ids)
# Embedding Layer
model = Embedding(input_dim=n_words, output_dim=word_embedding_size, input_length=max_len)(input)
# model = Embedding(input_dim=n_words, output_dim=word_embedding_size, input_length=140)(input)
# BI-LSTM Layer
model = Bidirectional(LSTM(units=word_embedding_size,
return_sequences=True,
dropout=0.5,
recurrent_dropout=0.5,
kernel_initializer=k.initializers.he_normal()))(model)
model = LSTM(units=word_embedding_size * 2,
return_sequences=True,
dropout=0.5,
recurrent_dropout=0.5,
kernel_initializer=k.initializers.he_normal())(model)
# TimeDistributed Layer
model = TimeDistributed(Dense(n_tags, activation="relu"))(model)
# CRF Layer
crf = CRF(n_tags)
out = crf(model) # output
model = Model(input, out)
# In[]
# model compile and fit
from keras.callbacks import ModelCheckpoint
import matplotlib.pyplot as plt
# Optimiser
adam = k.optimizers.Adam(lr=0.0005, beta_1=0.9, beta_2=0.999)
# Compile model
model.compile(optimizer=adam, loss=crf.loss_function, metrics=[crf.accuracy, 'accuracy'])
model.summary()
# Saving the best model only
filepath = "ner-bi-lstm-td-model-{val_accuracy:.2f}.hdf5"
checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max')
callbacks_list = [checkpoint]
# Fit the best model
history = model.fit(X_train, y_train, batch_size=256, epochs=10, validation_split=0.1, verbose=1,
callbacks=callbacks_list)
and here the code for the intent detection NN
#CNN architecture
from __future__ import print_function
import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras import backend as K
from keras import layers
batch_size = 128
epochs = 12
if nn_architecture == 'CNN':
model_CNN = Sequential()
e = Embedding(vocab_size, 300, weights=[embedding_matrix], input_length=max_length, trainable=False)
model_CNN.add(e)
model_CNN.add(Dropout(0.2))
# we add a Convolution1D, which will learn filters
# word group filters of size filter_length:
filters = 50
kernel_size = 3
hidden_dims = 250
model_CNN.add(layers.Conv1D(filters,
kernel_size,
padding='valid',
activation='relu',
strides=1))
# we use max pooling:
model_CNN.add(layers.GlobalMaxPooling1D())
# We add a vanilla hidden layer:
model_CNN.add(Dense(hidden_dims))
model_CNN.add(Dropout(0.2))
model_CNN.add(layers.Activation('relu'))
# We project onto a single unit output layer, and squash it with a sigmoid:
model_CNN.add(Dense(nbClasses)) # no_of_categories
model_CNN.add(layers.Activation('sigmoid'))
model_CNN.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
history_CNN =model_CNN.fit(X_train, Y_train_c,
batch_size=batch_size,
epochs=epochs,
#validation_split=0.2
)
# Epoch 12/12
# 38771/38771 [==============================] - 11s 276us/step -
#loss: 0.0046 - accuracy: 0.9985
What I would like to is to merge this two architectures in order to obtain this
Please help me ... thanks in advance
