In [1]:
!pip install gensim tensorflow wordcloud
!pip install -q tensorflow-hub
!pip install xgboost
!pip install keras
!pip install nltk
!pip install string
!pip install tqdm
import gzip
import gensim
import os
import sys
import json
import shutil
import time
import re
import tarfile
import zipfile
import numpy as np
import pandas as pd
import collections
import math
import random
import datetime as dt
import matplotlib.pyplot as plt
import seaborn as sns
%matplotlib inline
from wordcloud import WordCloud, STOPWORDS, ImageColorGenerator
import string
import nltk
nltk.download('stopwords')
from nltk.corpus import stopwords
nltk.download('punkt')
from nltk.tokenize import word_tokenize
from gensim.test.utils import get_tmpfile
from gensim.models import Word2Vec, FastText
from tqdm import tqdm
tqdm.pandas(desc="progress-bar")
from gensim.models import Doc2Vec
from gensim.models.doc2vec import LabeledSentence
import multiprocessing
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier
from sklearn.svm import SVC
from xgboost import XGBClassifier
from sklearn.decomposition import PCA
from sklearn.manifold import TSNE
from sklearn.pipeline import Pipeline
import tensorflow as tf
import tensorflow_hub as hub
from keras.preprocessing.text import Tokenizer
from keras.preprocessing.sequence import pad_sequences
from keras.models import Sequential
from keras.layers import Embedding,Dense,Flatten,GlobalMaxPooling1D,LSTM,Dropout, Activation,Bidirectional
from keras.layers.convolutional import Conv1D,MaxPooling1D
from keras.optimizers import Adam
from sklearn.model_selection import train_test_split
from sklearn.metrics import confusion_matrix, classification_report, accuracy_score, roc_curve, precision_recall_curve
from sklearn import utils
"""
We will ignore FutureWarning and DeprecationWarning
"""
import warnings
warnings.filterwarnings("ignore", category=FutureWarning)
warnings.filterwarnings("ignore", category=DeprecationWarning)
"""
We will ignore warnings
"""
warnings.filterwarnings("ignore")
if not sys.warnoptions:
warnings.simplefilter("ignore")
program_start_time=time.time()
Perform some housekeeping tasks¶
In [2]:
data_source_url = 'https://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz'
datasets = '/datasets/'
data_file_path = os.getcwd()+datasets+'aclImdb_v1.tar.gz'
data = 'data_'
MOVIE = 'movie'
data_folder = os.getcwd()+datasets+data+MOVIE
data_json_file = data+MOVIE+'.json'
def check_if_file_exists(file):
'''
Checks if 'file' exists
'''
try:
fh = open(file, 'r')
return True
except FileNotFoundError:
print('Please make sure file: ' + file + ' is present before continuing')
return False
def check_if_dir_exists(directory):
'''
Checks if 'directory' exists
'''
return(os.path.isdir(directory))
def store_json(write_this_data):
'''
Store json if we are processing the first time
'''
open(data_json_file, 'w').write(json.dumps(write_this_data))
def return_data_json():
'''
Return data json from data json file
'''
with open(data_json_file, encoding='utf-8') as data_file:
return json.loads(data_file.read())
Download data source
In [3]:
if not check_if_file_exists(data_file_path):
print('Start of data download')
wget.download(data_source_url, os.getcwd()+datasets)
print('Download complete')
else:
print('Data file already exists. Not downloading again!')
In [4]:
if not check_if_dir_exists(data_folder):
with tarfile.open(data_file_path) as tar:
tar.extractall(path=data_folder)
else:
print('Data foler exists. Won\'t copy again!')
In [5]:
data = {}
all_reviews = []
if not check_if_file_exists(data_json_file):
dirpath = data_folder+"/aclImdb/train/pos/"
for root, dirs, files in os.walk(dirpath):
for name in files:
full = os.path.join(root, name)
with open(full, 'r') as f:
string = f.read()
file_content = {"review": string,"rating": 1}
all_reviews.append(file_content)
dirpath = data_folder+"/aclImdb/train/neg/"
for root, dirs, files in os.walk(dirpath):
for name in files:
full = os.path.join(root, name)
with open(full, 'r') as f:
string = f.read()
file_content = {"review": string,"rating": 0}
all_reviews.append(file_content)
dirpath = data_folder+"/aclImdb/test/pos/"
for root, dirs, files in os.walk(dirpath):
for name in files:
full = os.path.join(root, name)
with open(full, 'r') as f:
string = f.read()
file_content = {"review": string,"rating": 1}
all_reviews.append(file_content)
dirpath = data_folder+"/aclImdb/test/neg/"
for root, dirs, files in os.walk(dirpath):
for name in files:
full = os.path.join(root, name)
with open(full, 'r') as f:
string = f.read()
file_content = {"review": string,"rating": 0}
all_reviews.append(file_content)
data[MOVIE] = all_reviews
store_json(data)
pre_loaded_data = return_data_json()
len(pre_loaded_data[MOVIE])
Out[5]:
Creating review dataframe and Data clean up
In [6]:
movie_df = pd.DataFrame(pre_loaded_data[MOVIE])
print('Before Cleanup : Shape of the Data Frame : {}'.format(movie_df.shape))
print('Remove missing values.')
movie_df.dropna(inplace=True)
movie_df.reset_index(drop=True,inplace=True)
print('Drop columns with duplicate data.')
movie_df.drop_duplicates()
print('After Cleanup : Shape of the Data Frame : {}'.format(movie_df.shape))
print('Counting null data per column.')
movie_df.isnull().sum()
Out[6]:
- The dataframe contains 50000 rows and 2 columns
Let us look at the data types of columns
In [7]:
movie_df.dtypes
Out[7]:
In [8]:
"""
Ratings
"""
movie_df.rating.unique()
Out[8]:
Let us explore the data a bit using head(), tail(), info(), describe()
In [9]:
movie_df.head()
Out[9]:
In [10]:
movie_df.tail()
Out[10]:
In [11]:
movie_df.info()
In [12]:
movie_df.describe()
Out[12]:
In [13]:
movie_df.describe(include='object')
Out[13]:
In [14]:
movie_df.describe(include='all')
Out[14]:
Creating a new column called "review_length" which is the length of the review column.
In [15]:
movie_df['review_length'] = movie_df['review'].apply(len)
movie_df.head()
Out[15]:
Data Visualization and Exploratory Data Analysis¶
Using FacetGrid from the seaborn library to create a grid of two histograms of review_length based off of the ratings
In [16]:
sns.set_style('darkgrid')
In [17]:
g = sns.FacetGrid(movie_df,col='rating',size=5)
g.map(plt.hist,'review_length',bins=50)
plt.show()
Let's try to explain why the x-axis goes all the way to 14000ish, this must mean that there is some really long message!
In [18]:
movie_df.review_length.describe()
Out[18]:
- Max review_length 13704.000000 means review is 13704 character long
Creating a boxplot of review_length for each rating category.
In [19]:
plt.figure(figsize=(10,8))
sns.boxplot(x='rating',y='review_length',data=movie_df,palette='rainbow')
plt.title("Boxplot of review length for each rating category.",fontsize=16)
plt.xlabel("Rating",fontsize=14)
plt.ylabel("Review Length",fontsize=14)
plt.show()
- Looking at the box and whisker plots for the review_length in words, we can see an exponential distribution. We can observe that the mass of the distribution can possibly be covered with 1400 to 1500 words.
Creating a countplot of the number of occurrences for each type of rating.
In [20]:
plt.figure(figsize=(10,8))
sns.countplot(x='rating',data=movie_df,palette='winter')
plt.title("Number of occurrences for each type of rating",fontsize=16)
plt.xlabel("Rating",fontsize=14)
plt.ylabel("Number of occurrences",fontsize=14)
plt.show()
Pre-processing of review text
In [21]:
def review_preprocess(review):
"""
Takes in a string of review, then performs the following:
1. Remove HTML tag from review
2. Remove URLs from review
3. Make entire review lowercase
4. Split the review in words
5. Remove all punctuation
6. Remove empty strings from review
7. Remove all stopwords
8. Returns a list of the cleaned review after jioning them back to a sentence
"""
en_stops = set(stopwords.words('english'))
"""
Removing HTML tag from review
"""
clean = re.compile('<.*?>')
review_without_tag = re.sub(clean, '', review)
"""
Removing URLs
"""
review_without_tag_and_url = re.sub(r"http\S+", "", review_without_tag)
review_without_tag_and_url = re.sub(r"www\S+", "", review_without_tag)
"""
Make entire string lowercase
"""
review_lowercase = review_without_tag_and_url.lower()
"""
Split string into words
"""
list_of_words = word_tokenize(review_lowercase)
"""
Remove punctuation
Checking characters to see if they are in punctuation
"""
list_of_words_without_punctuation=[''.join(this_char for this_char in this_string if (this_char in string.ascii_lowercase))for this_string in list_of_words]
"""
Remove empty strings
"""
list_of_words_without_punctuation = list(filter(None, list_of_words_without_punctuation))
"""
Remove any stopwords
"""
filtered_word_list = [w for w in list_of_words_without_punctuation if w not in en_stops]
"""
Returns a list of the cleaned review after jioning them back to a sentence
"""
return ' '.join(filtered_word_list)
In [22]:
"""
Here is the original reviews:
"""
movie_df['review'].tail()
Out[22]:
Applying pre-processing to reviews
In [23]:
start_time=time.time()
movie_df['review']=movie_df['review'].apply(review_preprocess)
print('Elapsed time for review preprocessing : ',((time.time()-start_time)/60),' in minutes')
In [24]:
"""
Here is the reviews after preprocessing :
"""
movie_df['review'].tail()
Out[24]:
The term frequency distribution of words in the review is obtained using nltk.FreqDist(). This provides us a rough idea of the main topic in the review dataset.¶
In [25]:
reviews = movie_df['review'].str.cat(sep=' ')
"""
function to split review into word
"""
tokens = word_tokenize(reviews)
vocabulary = set(tokens)
print('Number of vocabulary : {}'.format(len(vocabulary)))
frequency_distribution = nltk.FreqDist(tokens)
sorted(frequency_distribution,key=frequency_distribution.__getitem__, reverse=True)[0:50]
Out[25]:
Wordcloud visualization of frequent words¶
In [26]:
wordcloud = WordCloud().generate_from_frequencies(frequency_distribution)
plt.imshow(wordcloud)
plt.axis("off")
plt.show()
Train, test split¶
In [27]:
def split_train_test(x, y):
SEED = 2000
x_train, x_validation_and_test, y_train, y_validation_and_test = train_test_split(x, y, test_size=.2, random_state=SEED)
x_validation, x_test, y_validation, y_test = train_test_split(x_validation_and_test, y_validation_and_test, test_size=.5, random_state=SEED)
return x_train, y_train, x_test, y_test, x_validation, y_validation, pd.concat([x_train,x_validation,x_test])
In [28]:
X = movie_df.review
y = movie_df.rating
x_train, y_train, x_test, y_test, x_validation, y_validation, all_reviews = split_train_test(X, y)
We are done pre-processing! Now on to the data analysis and machine learning tasks...¶
- In the next code cell we reload data if needed or for first time load we simply read data into lists of tokenized items
- Then we train the Word2Vec model
Concerning embeddings:¶
- Developed by Tomas Mikolov in 2013 at Google, Word2Vec is one of the most popular algorithms to train "word embeddings" using a shallow two layer neural networks having one input layer, one hidden layer and one output layer. There are two models for generating word embeddings, i.e. CBOW and Skip-gram.
- Word Embedding is a language modeling technique that uses vectors with several dimensions to represent words from large amounts of unstructured text data. Word embeddings can be generated using various methods like neural networks, co-occurrence matrix, probabilistic models, etc.
- CBOW (Continuous Bag of Words) model CBOW model predicts the current word given a context of words. The input layer contains context words and the output layer contains current predicted word. The hidden layer contains the number of dimensions in which we want to represent current word at output layer. The CBOW architecture is shown below (image credit: Google)
- Skip-gram model flips CBOW's network architecture and aims to predict context given a word. Given current word, Skip gram predicts the surrounding context words. Input layer for it contains the current word and output layer provides the context words. The hidden layer consists of number of dimensions in which we want to represent current input word. The skip-gram architecture is shown below (image credit: Google)
- For our Word2Vec modeling we have used CBOW as it is faster and have better representation for more frequent words.
Reload previously processed model if exists
In [29]:
list_of_tokenized_reviews = []
skip_modeling = False
filename_to_save_model = MOVIE + ".model"
if check_if_file_exists(filename_to_save_model):
skip_modeling = True
In [30]:
if not skip_modeling:
for one_sentence in all_reviews:
list_of_tokenized_reviews.append(gensim.utils.simple_preprocess(one_sentence))
model = Word2Vec(list_of_tokenized_reviews, size=150, window=10, min_count=2, workers=10)
model.save(filename_to_save_model)
model = Word2Vec.load(filename_to_save_model)
else:
model = Word2Vec.load(filename_to_save_model)
Let's look at some output¶
In [31]:
"""
look up top 10 words similar to the word 'terrible'.
"""
w1 = "terrible"
model.wv.most_similar(positive=w1)
Out[31]:
In [32]:
"""
look up top 10 words similar to 'excellent'
"""
w1 = ["excellent"]
model.wv.most_similar (positive=w1)
Out[32]:
In [33]:
"""
look up top 3 words similar to 'movie'
"""
w1 = ["movie"]
model.wv.most_similar (positive=w1,topn=3)
Out[33]:
In [34]:
"""
look up top 5 words similar to 'worst'
"""
w1 = ["worst"]
model.wv.most_similar (positive=w1,topn=5)
Out[34]:
In [35]:
"""
similarity between two different words
"""
model.wv.similarity(w1="great",w2="worse")
Out[35]:
In [36]:
"""
similarity between two identical words
"""
model.wv.similarity(w1="outstanding",w2="outstanding")
Out[36]:
In [37]:
"""
similarity between two related words
"""
model.wv.similarity(w1="excellent",w2="outstanding")
Out[37]:
In [38]:
"""
Which one is the odd one out in this list?
"""
model.wv.doesnt_match(["best","great","good","disapointed"])
Out[38]:
In [39]:
"""
Which one is the odd one out in this list?
"""
model.wv.doesnt_match(["movie","film","show","book"])
Out[39]:
- Using dimensionality reduction algorithms like PCA and t-SNE to convert multi-dimensional word vectors to two dimensional plots
- The goal is to plot our 150 dimensions vectors into 2 dimensional graphs, and check if we can spot interesting patterns.
- Using PCA and t-SNE implementation from scikit-learn for dimension reductions.
- In the visualizations we will look at query word (in blue), and most similar words (in green), and list of words passed in the function (in red).
In [40]:
def word_vectors_plot(model, input_word, word_list):
"""
Seaborn plot results of query word and most similar words, alongwith other words in corpus
"""
word_arrays = np.empty((0, 150), dtype='f')
word_tags = [input_word]
color_list = ['blue']
"""
Creating Vector of query word
"""
word_arrays = np.append(word_arrays, model.wv.__getitem__([input_word]), axis=0)
"""
Find similar words
"""
similar_words = model.wv.most_similar([input_word],topn=8)
"""
Insert word vector for similar words into array
"""
for word_score in similar_words:
word_vector = model.wv.__getitem__([word_score[0]])
word_tags.append(word_score[0])
color_list.append('green')
word_arrays = np.append(word_arrays, word_vector, axis=0)
"""
Insert word vectors for other words into array
"""
for word in word_list:
word_vector = model.wv.__getitem__([word])
word_tags.append(word)
color_list.append('red')
word_arrays = np.append(word_arrays, word_vector, axis=0)
"""
Dimensionality from 150 to 17 dimensions with PCA
"""
reduce = PCA(n_components=17).fit_transform(word_arrays)
"""
Finds t-SNE coordinates for 2 dimensions
"""
np.set_printoptions(suppress=True)
Y = TSNE(n_components=2, random_state=0, perplexity=15).fit_transform(reduce)
"""
Sets everything up to plot
"""
df = pd.DataFrame({'x': [x for x in Y[:, 0]],
'y': [y for y in Y[:, 1]],
'words': word_tags,
'color': color_list})
fig, _ = plt.subplots()
fig.set_size_inches(9, 9)
"""
Original plot
"""
p1 = sns.regplot(data=df,
x="x",
y="y",
fit_reg=False,
marker="o",
scatter_kws={'s': 40,
'facecolors': df['color']
}
)
"""
Annotating word in plots
"""
for line in range(0, df.shape[0]):
p1.text(df["x"][line],
df['y'][line],
' ' + df["words"][line].title(),
horizontalalignment='left',
verticalalignment='bottom', size='medium',
color=df['color'][line],
weight='normal'
).set_size(15)
plt.xlim(Y[:, 0].min()-50, Y[:, 0].max()+50)
plt.ylim(Y[:, 1].min()-50, Y[:, 1].max()+50)
plt.title('t-SNE viz for input: {}'.format(input_word.title()),fontsize=16)
Eight Most Similar Words Vs. Eight Random Words¶
In [41]:
word_vectors_plot(model, 'action', ['good', 'performance', 'cool', 'life', 'issue', 'bore', 'buy', 'love'])
Eight Most Similar Words Vs. Ninth To Sixteenth Most Similar Words¶
In [42]:
word_vectors_plot(model, "excellent", [t[0] for t in model.wv.most_similar(positive=["excellent"],
topn=16)][8:])
Eight Most Similar Words Vs. Ninth To Sixteenth Most Similar Words¶
In [43]:
word_vectors_plot(model, "comedy", [t[0] for t in model.wv.most_similar(positive=["comedy"],
topn=16)][8:])
Eight Most Similar Words Vs. Ninth To Sixteenth Most Similar Words¶
In [44]:
word_vectors_plot(model, "story", [t[0] for t in model.wv.most_similar(positive=["story"],
topn=16)][8:])
Wordcloud visualization of positive words in reviews¶
In [45]:
pos_lst=[t[0] for t in model.wv.most_similar(positive=["outstanding"],topn=20)]
pos_wrd=' '.join(pos_lst)
print(pos_wrd)
In [46]:
wordcloud = WordCloud(background_color='white',
max_words=200000,
max_font_size=400).generate(pos_wrd)
plt.figure()
plt.imshow(wordcloud)
plt.axis("off")
plt.show()
Wordcloud visualization of negative words in reviews¶
In [47]:
neg_lst=[n[0] for n in model.wv.most_similar(positive=["awful"],topn=20)]
neg_wrd=' '.join(neg_lst)
print(neg_wrd)
In [48]:
wordcloud = WordCloud(background_color='black',
max_words=100,
max_font_size=50).generate(neg_wrd)
plt.figure()
plt.imshow(wordcloud)
plt.axis("off")
plt.show()
Generating Feature vectors¶
In [49]:
"""
Function to generate feature vectors
"""
def generate_feature_vectors(doc, model):
vec = np.zeros(150).reshape((1, 150))
count = 0
for word in gensim.utils.simple_preprocess(doc):
if model.__contains__(word.strip()):
count = count + 1
vec += model[word.strip()]
vec = vec / count
return vec
def generate_features(model, data):
features = np.concatenate([generate_feature_vectors(s, model) for s in data])
return features
"""
Generating train, test and validation vectors
"""
training_vectors = generate_features(model, x_train)
test_vectors = generate_features(model, x_test)
validation_vectors = generate_features(model, x_validation)
Word2Vec Word Embedding Based Sentiment Analysis using LogisticRegression¶
In [50]:
lr = LogisticRegression()
lr.fit(training_vectors, y_train)
print("***** Word2Vec Word Embedding Based Sentiment Analysis using LogisticRegression *******\n")
print("LogisticRegression Performance : \n")
print('Train-Set Score : {:.4f}'.format(lr.score(training_vectors, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,lr.predict(training_vectors))))
print("\nEvaluation on Validation-Set : ")
pred_val = lr.predict(validation_vectors)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(lr.score(validation_vectors, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = lr.predict(test_vectors)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(lr.score(test_vectors, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Word2Vec Word Embedding Based Sentiment Analysis using SVC¶
In [51]:
svm = SVC(kernel='linear')
svm.fit(training_vectors, y_train)
print("***** Word2Vec Word Embedding Based Sentiment Analysis using SVC *******\n")
print("SVC with linear kernel Performance : \n")
print('Train-Set Score : {:.4f}'.format(svm.score(training_vectors, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,svm.predict(training_vectors))))
print("\nEvaluation on Validation-Set : ")
pred_val = svm.predict(validation_vectors)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(svm.score(validation_vectors, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = svm.predict(test_vectors)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(svm.score(test_vectors, y_test)))
print("Test-Set Accuracy: {:.4f}".format(accuracy_score(y_test, pred)))
Word2Vec Word Embedding Based Sentiment Analysis using XGBClassifier¶
In [52]:
xgb = XGBClassifier()
xgb.fit(training_vectors, y_train)
print("***** Word2Vec Word Embedding Based Sentiment Analysis using XGBClassifier *******\n")
print("XGBClassifier Performance : \n")
print('Train-Set Score : {:.4f}'.format(xgb.score(training_vectors, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,xgb.predict(training_vectors))))
print("\nEvaluation on Validation-Set : ")
pred_val = xgb.predict(validation_vectors)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(xgb.score(validation_vectors, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = xgb.predict(test_vectors)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print("Test-Set Score : {:.4f}".format(xgb.score(test_vectors, y_test)))
print("Test-Set Accuracy: {:.4f}".format(accuracy_score(y_test, pred)))
Sentiment Analysis using Keras Convolutional Neural Networks(CNN)¶
In [53]:
"""
Create the tokenizer
"""
number_of_words=len(vocabulary)
tokenizer = Tokenizer(num_words=number_of_words)
"""
Fit the tokenizer
"""
tokenizer.fit_on_texts(x_train)
"""
Sequence encode
"""
X_token_train = tokenizer.texts_to_sequences(x_train)
X_token_test = tokenizer.texts_to_sequences(x_test)
X_token_validation = tokenizer.texts_to_sequences(x_validation)
"""
Adding 1 because of reserved 0 index
"""
vocabulary_size = len(tokenizer.word_index) + 1
print("x_train[2] : ",x_train[2])
print("\n X_token_train[2] : ",X_token_train[2])
print("\n vocab_size : ",vocabulary_size)
In [54]:
"""
Checking the index of each word by looking at the word_index dictionary of the Tokenizer object
"""
for word in ['famous','cartoon','studios', 'love', 'baby']:
print('{} : {}'.format(word, tokenizer.word_index[word]))
In [55]:
"""
Pad sequences
"""
max_length = 1500
X_token_train = pad_sequences(X_token_train, padding='post', maxlen=max_length)
X_token_test = pad_sequences(X_token_test, padding='post', maxlen=max_length)
X_token_validation = pad_sequences(X_token_validation, padding='post', maxlen=max_length)
In [56]:
%%time
"""
Create model
"""
embedding_dimension = 100
keras_cnn_model = Sequential()
keras_cnn_model.add(Embedding(input_dim=vocabulary_size,
output_dim=embedding_dimension,
input_length=max_length))
keras_cnn_model.add(Conv1D(128, 5, activation='relu'))
keras_cnn_model.add(GlobalMaxPooling1D())
keras_cnn_model.add(Dense(10, activation='relu'))
keras_cnn_model.add(Dense(1, activation='sigmoid'))
"""
Compile network
"""
keras_cnn_model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy'])
keras_cnn_model.summary()
In [57]:
%%time
"""
Fit network
"""
keras_cnn_model.fit(X_token_train, y_train,
epochs=5,
verbose=False,
validation_data=(X_token_validation, y_validation),
batch_size=10)
"""
Evaluate
"""
print("\n **** Sentiment Analysis Using Keras Convolutional Neural Networks(CNN) ****\n")
loss, accuracy = keras_cnn_model.evaluate(X_token_train, y_train, verbose=False)
print("Train-Set Accuracy: {:.4f}".format(accuracy))
print("\nEvaluation on Validation-Set : ")
pred_val=keras_cnn_model.predict_classes(X_token_validation)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
loss, accuracy = keras_cnn_model.evaluate(X_token_validation, y_validation, verbose=False)
print("Validation-Set Accuracy: {:.4f}".format(accuracy))
print("\nEvaluation on Test-Set : ")
pred=keras_cnn_model.predict_classes(X_token_test)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
loss, accuracy = keras_cnn_model.evaluate(X_token_test, y_test, verbose=False)
print("Test-Set Accuracy: {:.4f}".format(accuracy))
Sentiment Analysis Using Pre-trained Word2Vec Word Embedding To Keras CNN¶
In [58]:
"""
Vocabulary size
"""
num_of_words = list(model.wv.vocab)
print('Vocabulary size : %d' % len(num_of_words))
In [59]:
%%time
"""
Save model in ASCII
"""
file_name = 'movie_embedding_word2vec.txt'
model.wv.save_word2vec_format(file_name, binary=False)
In [60]:
"""
Load word embedding
"""
def load_word_embedding(file_name):
word_embedding = dict()
file = open(file_name,'r')
lines = file.readlines()[1:]
file.close()
"""
Mapping words to vectors
"""
for line in lines:
line_parts = line.split()
word_embedding[line_parts[0]] = np.asarray(line_parts[1:], dtype='float32')
return word_embedding
In [61]:
"""
Create a weight matrix for the Embedding layer
"""
def get_embedding_weight_matrix(wrd_embedding, vocabulary):
vocabulary_size = len(vocabulary) + 1
"""
Define weight matrix dimensions with all 0
"""
embedding_weight_matrix = np.zeros((vocabulary_size, 150))
"""
Step vocab, store vectors using the Tokenizer's integer mapping
"""
for wrd, i in vocabulary.items():
vector = wrd_embedding.get(wrd)
if vector is not None:
embedding_weight_matrix[i] = vector
return embedding_weight_matrix
In [62]:
"""
Load embedding from file
"""
raw_w2v_embedding = load_word_embedding('movie_embedding_word2vec.txt')
print('Completed creation of raw word2vec word embedding')
"""
Get weight vectors in the right order
"""
embedding_weight_vectors = get_embedding_weight_matrix(raw_w2v_embedding, tokenizer.word_index)
print('Completed creation of embedding weight vectors')
In [63]:
"""
Create the embedding layer
"""
embedding_layer = Embedding(vocabulary_size,
150,
weights=[embedding_weight_vectors],
input_length=max_length,
trainable=False)
print('Completed creation of embedding layer')
In [64]:
"""
Create model
"""
keras_cnn_w2v_model = Sequential()
keras_cnn_w2v_model.add(embedding_layer)
keras_cnn_w2v_model.add(Conv1D(filters=128, kernel_size=5, activation='relu'))
keras_cnn_w2v_model.add(MaxPooling1D(pool_size=2))
keras_cnn_w2v_model.add(Flatten())
keras_cnn_w2v_model.add(Dense(1, activation='sigmoid'))
keras_cnn_w2v_model.summary()
In [65]:
"""
Compile network
"""
keras_cnn_w2v_model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
"""
Fit network
"""
keras_cnn_w2v_model.fit(X_token_train, y_train,
epochs=5,
verbose=False,
validation_data=(X_token_validation, y_validation),
batch_size=10)
"""
Evaluate
"""
print("\n **** Sentiment Analysis Using Pre-trained Word2Vec Word Embedding To Keras CNN ****\n")
loss, accuracy = keras_cnn_w2v_model.evaluate(X_token_train, y_train, verbose=False)
print("Train-Set Accuracy: {:.4f}".format(accuracy))
print("\nEvaluation on Validation-Set : ")
pred_val=keras_cnn_w2v_model.predict_classes(X_token_validation)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
loss, accuracy = keras_cnn_w2v_model.evaluate(X_token_validation, y_validation, verbose=False)
print("Validation-Set Accuracy: {:.4f}".format(accuracy))
print("\nEvaluation on Test-Set : ")
pred=keras_cnn_w2v_model.predict_classes(X_token_test)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
loss, accuracy = keras_cnn_w2v_model.evaluate(X_token_test, y_test, verbose=False)
print("Test-Set Accuracy: {:.4f}".format(accuracy))
Sentiment Analysis Using Pre-trained Word2Vec Word Embedding To Keras CNN And Bidirectional LSTM¶
In [66]:
%%time
"""
Create model
"""
keras_cnn_bidir_lstm_w2v_model = Sequential()
keras_cnn_bidir_lstm_w2v_model.add(Embedding(vocabulary_size,
150,
weights=[embedding_weight_vectors],
input_length=max_length,
trainable=False))
keras_cnn_bidir_lstm_w2v_model.add(Conv1D(128, 5, activation='relu'))
keras_cnn_bidir_lstm_w2v_model.add(MaxPooling1D(pool_size=2))
keras_cnn_bidir_lstm_w2v_model.add(Bidirectional(LSTM(64)))
keras_cnn_bidir_lstm_w2v_model.add(Dropout(0.5))
keras_cnn_bidir_lstm_w2v_model.add(Dense(1, activation='sigmoid'))
"""
Compile network
"""
keras_cnn_bidir_lstm_w2v_model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
keras_cnn_bidir_lstm_w2v_model.summary()
In [67]:
%%time
"""
Fit train data
"""
keras_cnn_bidir_lstm_w2v_model.fit(X_token_train, y_train,
epochs=5,
verbose=False,
validation_data=(X_token_validation, y_validation),
batch_size=10)
In [68]:
"""
Evaluate
"""
print("**** Sentiment Analysis Using Pre-trained Word2Vec Word Embedding To Keras CNN And Bidirectional LSTM ****\n")
loss, accuracy = keras_cnn_bidir_lstm_w2v_model.evaluate(X_token_train, y_train, verbose=False)
print("Train-Set Accuracy: {:.4f}".format(accuracy))
print("\nEvaluation on Validation-Set : ")
pred_val=keras_cnn_bidir_lstm_w2v_model.predict_classes(X_token_validation)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
loss, accuracy = keras_cnn_bidir_lstm_w2v_model.evaluate(X_token_validation, y_validation, verbose=False)
print("Validation-Set Accuracy: {:.4f}".format(accuracy))
print("\nEvaluation on Test-Set : ")
pred=keras_cnn_bidir_lstm_w2v_model.predict_classes(X_token_test)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
loss, accuracy = keras_cnn_bidir_lstm_w2v_model.evaluate(X_token_test, y_test, verbose=False)
print("Test-Set Accuracy: {:.4f}".format(accuracy))
Sentiment Analysis Using Pre-trained Word2Vec Word Embedding To Keras Bidirectional LSTM¶
In [69]:
"""
Create model
"""
keras_bidir_lstm_w2v_model = Sequential()
keras_bidir_lstm_w2v_model.add(Embedding(vocabulary_size,
150,
weights=[embedding_weight_vectors],
input_length=max_length,
trainable=False))
keras_bidir_lstm_w2v_model.add(Bidirectional(LSTM(64)))
keras_bidir_lstm_w2v_model.add(Dropout(0.5))
keras_bidir_lstm_w2v_model.add(Dense(1, activation='sigmoid'))
"""
Compile network
"""
keras_bidir_lstm_w2v_model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
keras_bidir_lstm_w2v_model.summary()
In [70]:
%%time
"""
Fit train data
"""
keras_bidir_lstm_w2v_model.fit(X_token_train, y_train,
epochs=5,
verbose=False,
validation_data=(X_token_validation, y_validation),
batch_size=10)
In [71]:
"""
Evaluate
"""
print("**** Sentiment Analysis Using Pre-trained Word2Vec Word Embedding To Keras Bidirectional LSTM ****\n")
loss, accuracy = keras_bidir_lstm_w2v_model.evaluate(X_token_train, y_train, verbose=False)
print("Train-Set Accuracy: {:.4f}".format(accuracy))
print("\nEvaluation on Validation-Set : ")
pred_val=keras_bidir_lstm_w2v_model.predict_classes(X_token_validation)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
loss, accuracy = keras_bidir_lstm_w2v_model.evaluate(X_token_validation, y_validation, verbose=False)
print("Validation-Set Accuracy: {:.4f}".format(accuracy))
print("\nEvaluation on Test-Set : ")
pred=keras_bidir_lstm_w2v_model.predict_classes(X_token_test)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
loss, accuracy = keras_bidir_lstm_w2v_model.evaluate(X_token_test, y_test, verbose=False)
print("Test-Set Accuracy: {:.4f}".format(accuracy))
About Doc2Vec¶
- Doc2Vec is a generalization of the Word2Vec algorithm and applies at the document level. According to Mikolov et al. (2014), paragraphs in a document are mapped to a vector representation called paragraph vector. This is then combined with the word vectors by averaging or concatenating to predict the next word in a context. The paragraph vector is just like another word vector but it represents the missing context of the the topic of the paragraph.
- PV-DM or DM: is the Doc2Vec model analogous to CBOW in Word2Vec. The document vectors are obtained by training a neural network on the task of inferring a centre word based on context words and a context paragraph. See image below for architecture (image credit: Google)
- PV-DBOW or DBOW: is the Doc2Vec model analogous to Skip-gram in Word2Vec. The document vectors are obtained by training a neural network on the task of predicting a probability distribution of words in a paragraph given a randomly-sampled word from the document. See image below for architecture (image credit: Google)
Below we will see the usage gensim python library with comaprisons of each of these models and their combinations. That is:
- DBOW (Distributed Bag of Words)
- DMC (Distributed Memory Concatenated)
- DMM (Distributed Memory Mean)
- DBOW + DMC
- DBOW + DMM
In [72]:
"""
Function to labelize the reviews
"""
def labelize_review(reviews,label):
labelized_review = []
prefix = label
for indx, rvw in zip(reviews.index, reviews):
labelized_review.append(LabeledSentence(rvw.split(), [prefix + '_%s' % indx]))
return labelized_review
In [73]:
"""
labelize the reviews
"""
all_reviews_d2v = labelize_review(all_reviews, 'all')
Distributed Bag Of Words (DBOW)¶
In [74]:
%%time
"""
Create doc2vec DBOW model
"""
dbow_model = Doc2Vec(dm=0,
size=150,
negative=5,
min_count=2,
workers=10,
alpha=0.065,
min_alpha=0.065)
dbow_model.build_vocab([review for review in tqdm(all_reviews_d2v)])
In [75]:
%%time
"""
Train the model
"""
for epoch in range(3):
dbow_model.train(utils.shuffle([review for review in tqdm(all_reviews_d2v)]),
total_examples=len(all_reviews_d2v),
epochs=1)
dbow_model.alpha -= 0.002
dbow_model.min_alpha = dbow_model.alpha
In [76]:
"""
Function to generate vectors from corpus
"""
def generate_vectors(model, corpus, size):
vectors = np.zeros((len(corpus), size))
n = 0
for indx in corpus.index:
prefix = 'all_' + str(indx)
vectors[n] = model.docvecs[prefix]
n += 1
return vectors
In [77]:
"""
Generating train, test and validation vectors
"""
train_vectors_dbow = generate_vectors(dbow_model, x_train, 150)
test_vectors_dbow = generate_vectors(dbow_model, x_test, 150)
validation_vectors_dbow = generate_vectors(dbow_model, x_validation, 150)
Doc2Vec Distributed Bag Of Words(DBOW) Based Sentiment Analysis using LogisticRegression¶
In [78]:
logreg_dbow = LogisticRegression()
logreg_dbow.fit(train_vectors_dbow, y_train)
print("**** Doc2Vec Distributed Bag Of Words(DBOW) Based Sentiment Analysis using LogisticRegression ****\n")
print("LogisticRegression Performance : \n")
print('Train-Set Score : {:.4f}'.format(logreg_dbow.score(train_vectors_dbow, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,logreg_dbow.predict(train_vectors_dbow))))
print("\nEvaluation on Validation-Set : ")
pred_val = logreg_dbow.predict(validation_vectors_dbow)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(logreg_dbow.score(validation_vectors_dbow, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = logreg_dbow.predict(test_vectors_dbow)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(logreg_dbow.score(test_vectors_dbow, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Doc2Vec Distributed Bag Of Words(DBOW) Based Sentiment Analysis using SVC¶
In [79]:
svm_dbow = SVC(kernel='linear')
svm_dbow.fit(train_vectors_dbow, y_train)
print("**** Doc2Vec Distributed Bag Of Words(DBOW) Based Sentiment Analysis using SVC ****\n")
print("SVC with linear kernel Performance : \n")
print('Train-Set Score : {:.4f}'.format(svm_dbow.score(train_vectors_dbow, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,svm_dbow.predict(train_vectors_dbow))))
print("\nEvaluation on Validation-Set : ")
pred_val = svm_dbow.predict(validation_vectors_dbow)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(svm_dbow.score(validation_vectors_dbow, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = svm_dbow.predict(test_vectors_dbow)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(svm_dbow.score(test_vectors_dbow, y_test)))
print("Test-Set Accuracy: {:.4f}".format(accuracy_score(y_test, pred)))
Doc2Vec Distributed Bag Of Words(DBOW) Based Sentiment Analysis using XGBClassifier¶
In [80]:
xgb_dbow = XGBClassifier()
xgb_dbow.fit(train_vectors_dbow, y_train)
print("**** Doc2Vec Distributed Bag Of Words(DBOW) Based Sentiment Analysis using XGBClassifier ****\n")
print("XGBClassifier Performance : \n")
print('Train-Set Score : {:.4f}'.format(xgb_dbow.score(train_vectors_dbow, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,xgb_dbow.predict(train_vectors_dbow))))
print("\nEvaluation on Validation-Set : ")
pred_val = xgb_dbow.predict(validation_vectors_dbow)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(xgb_dbow.score(validation_vectors_dbow, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = xgb_dbow.predict(test_vectors_dbow)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print("Test-Set Score : {:.4f}".format(xgb_dbow.score(test_vectors_dbow, y_test)))
print("Test-Set Accuracy: {:.4f}".format(accuracy_score(y_test, pred)))
Distributed Momory (concatenated)¶
In [81]:
%%time
"""
Create Doc2Vec DMC model
"""
dmc_model = Doc2Vec(dm=1,
dm_concat=1,
size=150,
window=10,
negative=5,
min_count=2,
workers=10,
alpha=0.065,
min_alpha=0.065)
dmc_model.build_vocab([review for review in tqdm(all_reviews_d2v)])
In [82]:
%%time
"""
Train the model
"""
for epoch in range(3):
dmc_model.train(utils.shuffle([review for review in tqdm(all_reviews_d2v)]),
total_examples=len(all_reviews_d2v),
epochs=1)
dmc_model.alpha -= 0.002
dmc_model.min_alpha = dmc_model.alpha
In [83]:
"""
Generating train, test and validation vectors
"""
train_vectors_dmc = generate_vectors(dmc_model, x_train, 150)
test_vectors_dmc = generate_vectors(dmc_model, x_test, 150)
validation_vectors_dmc = generate_vectors(dmc_model, x_validation, 150)
Doc2Vec DMC Based Sentiment Analysis using LogisticRegression¶
In [84]:
logreg_dmc = LogisticRegression()
logreg_dmc.fit(train_vectors_dmc, y_train)
print("**** Doc2Vec Distributed Momory (concatenated) Based Sentiment Analysis using LogisticRegression ****\n")
print("LogisticRegression Performance : \n")
print('Train-Set Score : {:.4f}'.format(logreg_dmc.score(train_vectors_dmc, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,logreg_dmc.predict(train_vectors_dmc))))
print("\nEvaluation on Validation-Set : ")
pred_val = logreg_dmc.predict(validation_vectors_dmc)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(logreg_dmc.score(validation_vectors_dmc, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = logreg_dmc.predict(test_vectors_dmc)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(logreg_dmc.score(test_vectors_dmc, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Doc2Vec DMC Based Sentiment Analysis using SVC¶
In [85]:
svm_dmc = SVC(kernel='linear')
svm_dmc.fit(train_vectors_dmc, y_train)
print("**** Doc2Vec Distributed Momory (concatenated) Based Sentiment Analysis using SVC ****\n")
print("SVC with linear kernel Performance : \n")
print('Train-Set Score : {:.4f}'.format(svm_dmc.score(train_vectors_dmc, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,svm_dmc.predict(train_vectors_dmc))))
print("\nEvaluation on Validation-Set : ")
pred_val = svm_dmc.predict(validation_vectors_dmc)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(svm_dmc.score(validation_vectors_dmc, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = svm_dmc.predict(test_vectors_dmc)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(svm_dmc.score(test_vectors_dmc, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Doc2Vec DMC Based Sentiment Analysis using XGBClassifier¶
In [86]:
xgb_dmc = XGBClassifier()
xgb_dmc.fit(train_vectors_dmc, y_train)
print("**** Doc2Vec Distributed Momory (concatenated) Based Sentiment Analysis using XGBClassifier ****\n")
print("XGBClassifier Performance : \n")
print('Train-Set Score : {:.4f}'.format(xgb_dmc.score(train_vectors_dmc, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,xgb_dmc.predict(train_vectors_dmc))))
print("\nEvaluation on Validation-Set : ")
pred_val = xgb_dmc.predict(validation_vectors_dmc)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(xgb_dmc.score(validation_vectors_dmc, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = xgb_dmc.predict(test_vectors_dmc)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(xgb_dmc.score(test_vectors_dmc, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Distributed Memory (mean)¶
In [87]:
%%time
"""
Create doc2vec DMM model
"""
dmm_model = Doc2Vec(dm=1,
dm_mean=1,
size=150,
window=10,
negative=5,
min_count=2,
workers=10,
alpha=0.065,
min_alpha=0.065)
dmm_model.build_vocab([review for review in tqdm(all_reviews_d2v)])
In [88]:
%%time
"""
Train the model
"""
for epoch in range(3):
dmm_model.train(utils.shuffle([review for review in tqdm(all_reviews_d2v)]),
total_examples=len(all_reviews_d2v),
epochs=1)
dmm_model.alpha -= 0.002
dmm_model.min_alpha = dmm_model.alpha
In [89]:
"""
Generating train, test and validation vectors
"""
train_vectors_dmm = generate_vectors(dmm_model, x_train, 150)
test_vectors_dmm = generate_vectors(dmm_model, x_test, 150)
validation_vectors_dmm = generate_vectors(dmm_model, x_validation, 150)
Doc2Vec DMM Based Sentiment Analysis using LogisticRegression¶
In [90]:
logreg_dmm = LogisticRegression()
logreg_dmm.fit(train_vectors_dmm, y_train)
print("**** Doc2Vec Distributed Memory(mean) Based Sentiment Analysis using LogisticRegression ****\n")
print("LogisticRegression Performance : \n")
print('Train-Set Score : {:.4f}'.format(logreg_dmm.score(train_vectors_dmm, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,logreg_dmm.predict(train_vectors_dmm))))
print("\nEvaluation on Validation-Set : ")
pred_val = logreg_dmm.predict(validation_vectors_dmm)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(logreg_dmm.score(validation_vectors_dmm, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = logreg_dmm.predict(test_vectors_dmm)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(logreg_dmm.score(test_vectors_dmm, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Doc2Vec DMM Based Sentiment Analysis using SVC¶
In [91]:
svm_dmm = SVC(kernel='linear')
svm_dmm.fit(train_vectors_dmm, y_train)
print("**** Doc2Vec Distributed Memory(mean) Based Sentiment Analysis using SVC ****\n")
print("SVC With Linear Kernel Performance : \n")
print('Train-Set Score : {:.4f}'.format(svm_dmm.score(train_vectors_dmm, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,svm_dmm.predict(train_vectors_dmm))))
print("\nEvaluation on Validation-Set : ")
pred_val = svm_dmm.predict(validation_vectors_dmm)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(svm_dmm.score(validation_vectors_dmm, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = svm_dmm.predict(test_vectors_dmm)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(svm_dmm.score(test_vectors_dmm, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Doc2Vec DMM Based Sentiment Analysis using XGBClassifier¶
In [92]:
xgb_dmm = XGBClassifier()
xgb_dmm.fit(train_vectors_dmm, y_train)
print("**** Doc2Vec Distributed Memory(mean) Based Sentiment Analysis using XGBClassifier ****\n")
print("XGBClassifier Performance : \n")
print('Train-Set Score : {:.4f}'.format(xgb_dmm.score(train_vectors_dmm, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,xgb_dmm.predict(train_vectors_dmm))))
print("\nEvaluation on Validation-Set : ")
pred_val = xgb_dmm.predict(validation_vectors_dmm)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(xgb_dmm.score(validation_vectors_dmm, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = xgb_dmm.predict(test_vectors_dmm)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(xgb_dmm.score(test_vectors_dmm, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Combination of Doc2Vec Distributed Bag Of Words (DBOW) And Distributed Memory(Concatenated)¶
Now, I have the document vectors from three different models, now I can concatenate them in combination to see how it affects the performance. Below I defined a function to concatenate document vectors from different models
In [93]:
"""
Function to concatenate document vectors from different models
"""
def generate_concat_vectors(model1,model2, corpus, size):
vectors = np.zeros((len(corpus), size))
n = 0
for indx in corpus.index:
prefix = 'all_' + str(indx)
vectors[n] = np.append(model1.docvecs[prefix],model2.docvecs[prefix])
n += 1
return vectors
In [94]:
"""
Generating train, test and validation document vectors
"""
train_vectors_dbow_dmc = generate_concat_vectors(dbow_model,dmc_model, x_train, 300)
test_vectors_dbow_dmc = generate_concat_vectors(dbow_model,dmc_model, x_test, 300)
validation_vectors_dbow_dmc = generate_concat_vectors(dbow_model,dmc_model, x_validation, 300)
Combination of Doc2Vec DBOW And DMC Based Sentiment Analysis using LogisticRegression¶
In [95]:
logreg_dbow_dmc = LogisticRegression()
logreg_dbow_dmc.fit(train_vectors_dbow_dmc, y_train)
print("**** Combination of Doc2Vec DBOW And DMC Based Sentiment Analysis using LogisticRegression ****\n")
print("LogisticRegression Performance : \n")
print('Train-Set Score : {:.4f}'.format(logreg_dbow_dmc.score(train_vectors_dbow_dmc, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,logreg_dbow_dmc.predict(train_vectors_dbow_dmc))))
print("\nEvaluation on Validation-Set : ")
pred_val = logreg_dbow_dmc.predict(validation_vectors_dbow_dmc)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(logreg_dbow_dmc.score(validation_vectors_dbow_dmc, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = logreg_dbow_dmc.predict(test_vectors_dbow_dmc)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(logreg_dbow_dmc.score(test_vectors_dbow_dmc, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Combination of Doc2Vec DBOW And DMC Based Sentiment Analysis using SVC¶
In [96]:
svm_dbow_dmc = SVC(kernel='linear')
svm_dbow_dmc.fit(train_vectors_dbow_dmc, y_train)
print("**** Combination of Doc2Vec DBOW And DMC Based Sentiment Analysis using SVC ****\n")
print("SVC with linear kernel Performance : \n")
print('Train-Set Score : {:.4f}'.format(svm_dbow_dmc.score(train_vectors_dbow_dmc, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,svm_dbow_dmc.predict(train_vectors_dbow_dmc))))
print("\nEvaluation on Validation-Set : ")
pred_val = svm_dbow_dmc.predict(validation_vectors_dbow_dmc)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(svm_dbow_dmc.score(validation_vectors_dbow_dmc, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = svm_dbow_dmc.predict(test_vectors_dbow_dmc)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(svm_dbow_dmc.score(test_vectors_dbow_dmc, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Combination of Doc2Vec DBOW And DMC Based Sentiment Analysis using XGBClassifier¶
In [97]:
xgb_dbow_dmc =XGBClassifier()
xgb_dbow_dmc.fit(train_vectors_dbow_dmc, y_train)
print("**** Combination of Doc2Vec DBOW And DMC Based Sentiment Analysis using XGBClassifier ****\n")
print("XGBClassifier Performance : \n")
print('Train-Set Score : {:.4f}'.format(xgb_dbow_dmc.score(train_vectors_dbow_dmc, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,xgb_dbow_dmc.predict(train_vectors_dbow_dmc))))
print("\nEvaluation on Validation-Set : ")
pred_val = xgb_dbow_dmc.predict(validation_vectors_dbow_dmc)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(xgb_dbow_dmc.score(validation_vectors_dbow_dmc, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = xgb_dbow_dmc.predict(test_vectors_dbow_dmc)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(xgb_dbow_dmc.score(test_vectors_dbow_dmc, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Combination of Doc2Vec Distributed Bag Of Words (DBOW) And Distributed Memory(Mean)¶
In [98]:
"""
Generating train, test and validation document vectors
"""
train_vectors_dbow_dmm = generate_concat_vectors(dbow_model,dmm_model, x_train, 300)
test_vectors_dbow_dmm = generate_concat_vectors(dbow_model,dmm_model, x_test, 300)
validation_vectors_dbow_dmm = generate_concat_vectors(dbow_model,dmm_model, x_validation, 300)
Combination of Doc2Vec DBOW And DMM Based Sentiment Analysis using LogisticRegression¶
In [99]:
logreg_dbow_dmm = LogisticRegression()
logreg_dbow_dmm.fit(train_vectors_dbow_dmm, y_train)
print("**** Combination of Doc2Vec DBOW And DMM Based Sentiment Analysis using LogisticRegression ****\n")
print("LogisticRegression Performance : \n")
print('Train-Set Score : {:.4f}'.format(logreg_dbow_dmm.score(train_vectors_dbow_dmm, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,logreg_dbow_dmm.predict(train_vectors_dbow_dmm))))
print("\nEvaluation on Validation-Set : ")
pred_val = logreg_dbow_dmm.predict(validation_vectors_dbow_dmm)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(logreg_dbow_dmm.score(validation_vectors_dbow_dmm, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = logreg_dbow_dmm.predict(test_vectors_dbow_dmm)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(logreg_dbow_dmm.score(test_vectors_dbow_dmm, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Combination of Doc2Vec DBOW And DMM Based Sentiment Analysis using SVC¶
In [100]:
svm_dbow_dmm = SVC(kernel='linear')
svm_dbow_dmm.fit(train_vectors_dbow_dmm, y_train)
print("**** Combination of Doc2Vec DBOW And DMM Based Sentiment Analysis using SVC ****\n")
print("SVC with linear kernel Performance : \n")
print('Train-Set Score : {:.4f}'.format(svm_dbow_dmm.score(train_vectors_dbow_dmm, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,svm_dbow_dmm.predict(train_vectors_dbow_dmm))))
print("\nEvaluation on Validation-Set : ")
pred_val = svm_dbow_dmm.predict(validation_vectors_dbow_dmm)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(svm_dbow_dmm.score(validation_vectors_dbow_dmm, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = svm_dbow_dmm.predict(test_vectors_dbow_dmm)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(svm_dbow_dmm.score(test_vectors_dbow_dmm, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Combination of Doc2Vec DBOW And DMM Based Sentiment Analysis using XGBClassifier¶
In [101]:
xgb_dbow_dmm =XGBClassifier()
xgb_dbow_dmm.fit(train_vectors_dbow_dmm, y_train)
print("**** Combination of Doc2Vec DBOW And DMM Based Sentiment Analysis using XGBClassifier ****\n")
print("XGBClassifier Performance : \n")
print('Train-Set Score : {:.4f}'.format(xgb_dbow_dmm.score(train_vectors_dbow_dmm, y_train)))
print('Train-Set Accuracy : {:.4f}'.format(accuracy_score(y_train,xgb_dbow_dmm.predict(train_vectors_dbow_dmm))))
print("\nEvaluation on Validation-Set : ")
pred_val = xgb_dbow_dmm.predict(validation_vectors_dbow_dmm)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
print('Validation-Set Score : {:.4f}'.format(xgb_dbow_dmm.score(validation_vectors_dbow_dmm, y_validation)))
print('Validation-Set Accuracy:{:.4f}'.format(accuracy_score(y_validation, pred_val)))
print("\nEvaluation on Test-Set : ")
pred = xgb_dbow_dmm.predict(test_vectors_dbow_dmm)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
print('Test-Set Score : {:.4f}'.format(xgb_dbow_dmm.score(test_vectors_dbow_dmm, y_test)))
print('Test-Set Accuracy:{:.4f}'.format(accuracy_score(y_test, pred)))
Sentiment Analysis Using Combination of Doc2Vec DBOW And DMC Document Embedding and Keras Neural Network¶
In [102]:
"""
Create model
"""
keras_d2v_combo_dbow_dmc_model = Sequential()
keras_d2v_combo_dbow_dmc_model.add(Dense(128, activation='relu', input_dim=300))
keras_d2v_combo_dbow_dmc_model.add(Dense(128, activation='relu'))
keras_d2v_combo_dbow_dmc_model.add(Dense(1, activation='sigmoid'))
"""
Compile network
"""
keras_d2v_combo_dbow_dmc_model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy'])
keras_d2v_combo_dbow_dmc_model.summary()
In [103]:
"""
Fit network
"""
keras_d2v_combo_dbow_dmc_model.fit(train_vectors_dbow_dmc, y_train,
validation_data=(validation_vectors_dbow_dmc, y_validation),
epochs=5,
batch_size=10,
verbose=False)
"""
Evaluate
"""
print("**** Sentiment Analysis Using Combination of Doc2Vec DBOW And DMC Document Embedding and Keras Neural Network ****\n")
loss, accuracy = keras_d2v_combo_dbow_dmc_model.evaluate(train_vectors_dbow_dmc, y_train, verbose=False)
print("Train-Set Accuracy: {:.4f}".format(accuracy))
print("\nEvaluation on Validation-Set : ")
pred_val=keras_d2v_combo_dbow_dmc_model.predict_classes(validation_vectors_dbow_dmc)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
loss, accuracy = keras_d2v_combo_dbow_dmc_model.evaluate(validation_vectors_dbow_dmc, y_validation, verbose=False)
print("Validation-Set Accuracy: {:.4f}".format(accuracy))
print("\nEvaluation on Test-Set : ")
pred=keras_d2v_combo_dbow_dmc_model.predict_classes(test_vectors_dbow_dmc)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
loss, accuracy = keras_d2v_combo_dbow_dmc_model.evaluate(test_vectors_dbow_dmc, y_test, verbose=False)
print("Test-Set Accuracy: {:.4f}".format(accuracy))
Sentiment Analysis Using Combination of Doc2Vec DBOW And DMM Document Embedding and Keras Neural Network¶
In [104]:
"""
Create model
"""
keras_d2v_combo_dbow_dmm_model = Sequential()
keras_d2v_combo_dbow_dmm_model.add(Dense(128, activation='relu', input_dim=300))
keras_d2v_combo_dbow_dmm_model.add(Dense(128, activation='relu'))
keras_d2v_combo_dbow_dmm_model.add(Dense(1, activation='sigmoid'))
"""
Compile network
"""
keras_d2v_combo_dbow_dmm_model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy'])
keras_d2v_combo_dbow_dmm_model.summary()
In [105]:
"""
Fit network
"""
keras_d2v_combo_dbow_dmm_model.fit(train_vectors_dbow_dmm, y_train,
validation_data=(validation_vectors_dbow_dmm, y_validation),
epochs=5,
batch_size=10,
verbose=False)
"""
Evaluate
"""
print("**** Sentiment Analysis Using Combination of Doc2Vec DBOW And DMM Document Embedding and Keras Neural Network ****\n")
loss, accuracy = keras_d2v_combo_dbow_dmm_model.evaluate(train_vectors_dbow_dmm, y_train, verbose=False)
print("Train-Set Accuracy: {:.4f}".format(accuracy))
print("\nEvaluation on Validation-Set : ")
pred_val=keras_d2v_combo_dbow_dmm_model.predict_classes(validation_vectors_dbow_dmm)
print("Classification report:\n {}".format(classification_report(y_validation, pred_val)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_validation, pred_val)))
loss, accuracy = keras_d2v_combo_dbow_dmm_model.evaluate(validation_vectors_dbow_dmm, y_validation, verbose=False)
print("Validation-Set Accuracy: {:.4f}".format(accuracy))
print("\nEvaluation on Test-Set : ")
pred=keras_d2v_combo_dbow_dmm_model.predict_classes(test_vectors_dbow_dmm)
print("Classification report:\n {}".format(classification_report(y_test, pred)))
print("Confusion matrix:\n {}".format(confusion_matrix(y_test, pred)))
loss, accuracy = keras_d2v_combo_dbow_dmm_model.evaluate(test_vectors_dbow_dmm, y_test, verbose=False)
print("Test-Set Accuracy: {:.4f}".format(accuracy))
In [106]:
print("Time elapsed : ",(round(((time.time()-program_start_time)/3600),2))," in hours")