Preprocessing of raw data in NLP: Tokenization, Stemming, Lemmatization and Vectorization

NLP (Natural Language Processing) is a very interesting branch of Artificial Intelligence. Natural language is a language which we human use to communicate and interact with each other. In NLP, we are teaching computers to understand, interpret and manipulate human languages. In this article, we will focus on some of the preprocessing tasks which we perform on the raw data like Tokenization, Stemming, Lemmatization and Vectorization.

While processing a natural language which we human speak, we need to take care of following things:

1. Syntax: The sentence should be grammatically correct. The arrangement of words in a sentence should follow all the grammar rules defined by a language. 

2. Semantics: It deals with the meaning of words and their interpretation within sentences.

3. Pragmatics: Same as semantics but it also consider context in which the word is used.

Applications of NLP

Applications of NLP (Natural Language Processing) are unlimited. I have listed few of those:

1. Machine translation (like Google Translate)
2. Sentiment analysis (reviews and comments on e-commerce and social-networking sites)
3. Text classification, generation and automatic summarization

4. Automated question answering and conversational interfaces (like chatbots)
5. Personal assistants (like Alexa, Siri, Google Assistant, Cortana etc.)

6. Auto-correct grammatical mistakes (MS Word and Grammarly use NLP to check grammatical errors)

7. Spam filtering, auto-complete, auto-correct, auto-tagging, topic modelling, sentence segmentation, speech recognition, part of speech tagging, named entity recognition, duplicates detection and a lot more...

NLP Toolkit (library) in Python

There are a lot of libraries in Python for NLP but the most commonly used library is NLTK( Natural Language Toolkit). It provides very efficient modules for preprocessing and cleaning of raw data like removing punctuation, tokenizing, removing stopwords, stemming, lemmatization, vectorization, tagging, parsing, and more.

Pre-processing of raw data in NLP

Following are the basic steps which we need to perform while cleaning the raw data in NLP:

1. Remove Punctuation
2. Tokenization
3. Remove Stopwords

4. Stemming / Lemmatization
5. Vectorization

1. Remove Punctuation: First of all, we should remove all the punctuation marks (like comma, semicolon, colon, apostrophe, quotation marks, dash, hyphen, brackets, braces, parentheses, ellipsis etc.) from the text as these carry negligible weight.

2. Tokenization: Now create a list of words used in the text. Each word is called a token. We can use regular expression to find out tokens from the sentences otherwise NLTK has efficient modules for this task.

3. Remove Stopwords: Now we need to remove all the stopwords from the token list. Stopwords are the words which occur frequently in a sentence but carry little weight (like the, for, is, and, or, been, to, this, that, but, if, in, a, as etc.).

4.1 Stemming: It is used to reduce the number of tokens just like removing stopwords. In this process, we reduce inflected words to their word stem or root. We keep only the semantic meaning of similar words.

Examples: 

1) Tokens like stemming and stemmed are converted to a token stem.

2) Tokens like working, worked, works and work are converted to a token work.

Points 1 and 2 clearly illustrate that how can we reduce the number of tokens in a token list using stemming. But wait! There is a problem. Consider following examples of stemming:

3) Tokens like meanness and meaning are converted to a token mean. Now this is wrong. Both tokens have different meanings, even then its treating both as same.

4) Tokens like goose and geese are converted to the tokens goos and gees respectively (it will just remove "e" suffix from both the tokens). Now this is again wrong. "geese" is just a plural of "goose", even then its treating both tokens as different.

Points 3 and 4 can be resolved using Lemmatization.

NLTK library has 4 stemmers:

1) Porter Stemmer
2) Snowball Stemmer
3) Lancaster Stemmer
4) Regex-based Stemmer

I mainly use Porter stemmer for stemming the tokens in my NLP code.

4.2: Lemmatization: We saw the limitation of stemming in above examples (3 and 4). We can overcome these limitations using Lemmatization. It is more powerful and sophisticated as compared to stemming and returns more accurate and meaningful words / tokens by considering the context in which the word is used in a sentence.

But the tradeoff is that, it is slower and complex as compared to the stemming.

Examples: 

1) Tokens like meanness and meaning are retained as it is instead of reducing it to mean (unlike stemming).

2) Tokens like goose and geese are converted to a token goose which is right. We should get rid of the token "geese" as it is just a plural of "goose".

I mainly use WordNet Lemmatizer present in NLTK library.

5. Vectorization: Machine Learning algorithms don't understand text. These need numeric data for matrix multiplications. Till now, we have just cleaned our tokens. So, in this process, we encode our final tokens into numbers to create feature vectors so that algorithms can understand. In other words, we will fit and transform vectorization methods to our preprocessed and cleaned data which we created till lemmatization.

Document-term matrix: Let's first understand this term before proceeding further. We use document term matrix to represent the words in the text in the form of matrix of numbers. The rows of the matrix represent the text responses to be analyzed, and the columns of the matrix represent the words / tokens from the text that are to be used in the analysis.

Types of Vectorization

There are mainly 3 types vectorization:

1) Count vectorization
2) N-grams vectorization
3) Term Frequency - Inverse Document Frequency (TF-IDF)

1) Count vectorization: It creates a document-term matrix which contains the count of each unique word / token in the text response.

2) N-grams vectorization: It creates a document-term matrix which also considers context of the word depending upon the value of N.

If N = 2, it is called bi-gram,
If N = 3, it is called tri-gram,
If N = 4, it is called four-gram and so on...

We need to be careful about value of N and choose it wisely.

Example: Consider a sentence "NLP is awesome". Count vectorization will create a column corresponding to each word in document-term matrix while N-gram will create columns like following in case of bi-gram:

"NLP is", "is awesome"

3) Term Frequency - Inverse Document Frequency (TF-IDF) - It is just like count vectorization but instead of count, it stores weightage of each word by using following  formula:

w(i, j) = weightage of a particular word "i" in a document "j"

tf(i, j) = term frequency of a word "i" in document "j" i.e. number of times the word "i" occurs in a document "j" divided by total number of words in document "j"

N = number of total documents

df(i) = number of documents containing the word "i"

So, in this way, TF-IDF considers two facts while calculating the weightage of a word or token: 

1) how frequent the word occurs in a particular document 

2) and how frequent that word occurs in other documents

Example: Consider that we have 10 text messages and one of the text messages is "NLP is awesome". No other message contains the word "NLP". Now lets calculate weightage of the word NLP.

tf(i, j) = number of times the word NLP occurs in the text message divided by the total number of words in the text message. It comes out to be (1/3) as there are three words and NLP occurs only one time.

N = 10 as there are 10 text messages. 

df(i) = number of text messages containing the word NLP which in our case is 1.

So, the final equation becomes:

Weightage of NLP = (1/3) * log(10/1)

In this way, we fill all the rows and column of document-term matrix in TF-IDF.

Image Recognition: Text Detection (Optical Character Recognition) using Google Cloud Vision API

Google Cloud Vision API helps in label detection, face detection, logo detection, landmark detection and text detection (OCR: Optical Character Recognition). In this article, we will see how can we use Google Cloud Vision API to extract the text from the image? This is a step by step guide for text detection (OCR) using Google Cloud Vision API. Let's follow it.

I will directly start from step 5. First 4 steps are same as mentioned in my previous post on label detection using Google Cloud Vision API.

You can download my Jupyter notebook containing below code from here.

 
Step 5: Import required libraries

from googleapiclient.discovery import build
from oauth2client.client import GoogleCredentials
from base64 import b64encode

You may get import error "no module name..." if you have not already installed Google API Python client. Use following command to install it.

pip install --upgrade google-api-python-client

If you also get import error for oauth2client, you must install it using following command:

pip3 install --upgrade oauth2client

Step 6: Load credentials file

Load the credentials file (which we created in step 3 of my previous article) and create a service object using it.

CREDENTIAL_FILE = 'credentials.json'
credentials = GoogleCredentials.from_stream(CREDENTIAL_FILE)
service = build('vision', 'v1', credentials=credentials)

Step 7: Load image file (from which we need to extract the text)

I will load an image of cover page of my deep learning book and encode it so that it becomes compatible with the cloud vision API.

IMAGE_FILE = book_cover_page.jpg'

with open(IMAGE_FILE, 'rb') as file:
    image_data = file.read()
    encoded_image_data = b64encode(image_data).decode('UTF-8')

Step 8: Create a batch request

We will create a batch request which we will send to the cloud vision API. In the batch request, we will include the above encoded image and the instruction as TEXT_DETECTION.

batch_request = [{
    'image':{'content':encoded_image_data},
    'features':[{'type':'TEXT_DETECTION'}],
}]

Step 9: Create a request

request = service.images().annotate(body={'requests':batch_request})

Step 10: Execute the request

response = request.execute()

This step will throw an error if you have not enabled billing (as mentioned in step 4 of my previous article). So, you must enable the billing in order to use Google Cloud Vision API. The charges are very reasonable. So, don't think too much and provide credit card details. For me, Google charged INR 1 and then refunded it back.

Step 11: Process the response

For error handling, include this code:

if 'error' in response:
    raise RuntimeError(response['error'])

We are interested in text annotations here. So, fetch it from the response and display the results.

labels = response['responses'][0]['textAnnotations']

extracted_text = extracted_texts[0]
print(extracted_text['description'], extracted_text['boundingPoly'])

Output:

Objective Type Questions and Answers in Deep Learning
Deep
Learning
ARTIFICIAL
INTELLIGENCE
MACHINE
LEARNING
DEEP
LEARNING
NARESH KUMAR
 {'vertices': [{'x': 42, 'y': 77}, {'x': 2365, 'y': 77}, {'x': 2365, 'y': 3523}, {'x': 42, 'y': 3523}]}

You can test the above code using different images and check the accuracy of the API.

Image Recognition: Label Detection using Google Cloud Vision API

Google Cloud Vision API helps in label detection, face detection, logo detection, landmark detection and text detection. In this article, we will see how can we use Google Cloud Vision API to identify labels in the image? This is a step by step guide for label detection using Google Cloud Vision API. Let's follow it.

Step 1: Setup a Google Cloud Account

A) Go to: https://console.cloud.google.com/
B) Login with your google credentials
C) You will see a dashboard. Create a Project if not already created.

Step 2: Enable Cloud Vision API

A) Go to console
B) Click on Navigation Menu
C) Click on API & Services >> Library
D) Search "cloud vision" and you will get the "Cloud Vision API". Enable this API if not already enabled.

Step 3: Download credentials file

A) Go to console
B) Click on Navigation Menu
C) Click on API & Services >> Credentials
D) Click on Create Credentials dropdown >> Service account key >> New service account
E) Enter Service account name
F) Select any role. I had selected Project >> Viewer
G) Save the file as JSON on your hard drive. Rename it to 'credentials.json'.

Step 4: Add billing information

A) Go to console
B) Click on Navigation Menu
C) Click on Billing

Now open the Jupyter notebook and try using this API. You can download my Jupyter notebook containing below code from here.
 
Step 5: Import required libraries

from googleapiclient.discovery import build
from oauth2client.client import GoogleCredentials
from base64 import b64encode

You may get import error "no module name..." if you have not already installed Google API Python client. Use following command to install it.

pip install --upgrade google-api-python-client

If you also get import error for oauth2client, you must install it using following command:

pip3 install --upgrade oauth2client

Step 6: Load credentials file

Load the credentials file (which we created in step 3) and create a service object using it.

CREDENTIAL_FILE = 'credentials.json'
credentials = GoogleCredentials.from_stream(CREDENTIAL_FILE)
service = build('vision', 'v1', credentials=credentials)

Step 7: Load image file (which needs to be tested)

We will load an image of a cat and encode it so that it becomes compatible with the cloud vision API.

IMAGE_FILE = 'cat.jpg'
with open(IMAGE_FILE, 'rb') as file:
    image_data = file.read()
    encoded_image_data = b64encode(image_data).decode('UTF-8')

Step 8: Create a batch request

We will create a batch request which we will send to the cloud vision API. In the batch request, we will include the above encoded image and the instruction as LABEL_DETECTION.

batch_request = [{
    'image':{'content':encoded_image_data},
    'features':[{'type':'LABEL_DETECTION'}],
}]

Step 9: Create a request

request = service.images().annotate(body={'requests':batch_request})

Step 10: Execute the request

response = request.execute()

This step will throw an error if you have not enabled billing (as mentioned in step 4). So, you must enable the billing in order to use Google Cloud Vision API. The charges are very reasonable. So, don't think too much and provide credit card details. For me, Google charged INR 1 and then refunded it back.

Step 11: Process the response

For error handling, include this code:

if 'error' in response:
    raise RuntimeError(response['error'])

We are interested in label annotations here. So, fetch it from the response and display the results.

labels = response['responses'][0]['labelAnnotations']

for label in labels:
    print(label['description'], label['score'])

Output:

Cat 0.99598557
Mammal 0.9890478
Vertebrate 0.9851104
Small to medium-sized cats 0.978553
Felidae 0.96784574
European shorthair 0.960582
Tabby cat 0.9573447
Whiskers 0.9441685
Dragon li 0.93990624
Carnivore 0.9342105

You can test the above code using different images and check the accuracy of the API.

Image Recognition using Pre-trained VGG16 model in Keras

Lets use a pre-trained VGG16 model to predict an image from ImageNet database. We will load an image, convert that image to numpy array, preprocess that array and let the pre-trained VGG16 model predict the image.

VGG16 is a CNN model. To know more about CNN, you can visit my this post. We are not fine-tuning the VGG16 model here. We are using it as it is. To fine-tune the existing VGG16 model, you can visit my this post.

You can download my Jupyter notebook containing following code from here.

Step 1: Import required libraries

import numpy as np
from keras.applications import vgg16
from keras.preprocessing import image

Step 2: Load pre-trained weights from VGG16 model for ImageNet dataset

model = vgg16.VGG16(weights='imagenet')

Step 3: Load image to predict

img = image.load_img('cat.jpg', target_size=(224, 224))
img

Please note that we need to reshape the image to 224X224 as it is a requirement for VGG16 model. You can download this image from ImageNet official website.

Step 4: Convert the image into numpy array

arr = image.img_to_array(img)
arr.shape

(224, 224, 3)

Step 5: Expand the array dimension

arr = np.expand_dims(arr, axis=0)
arr.shape

(1, 224, 224, 3)

Step 6: Preprocess the array

arr = vgg16.preprocess_input(arr)
arr

Step 7: Predict from the model

predictions = model.predict(arr)

predictions

We get an array as an output which is hard to understand. So, lets simplify it and see top 5 predictions made by the VGG16 model.

vgg16.decode_predictions(predictions, top=5)

[[('n02123045', 'tabby', 0.7138179),
  ('n02123159', 'tiger_cat', 0.21695374),
  ('n02124075', 'Egyptian_cat', 0.043560617),
  ('n04040759', 'radiator', 0.0053847637),
  ('n04553703', 'washbasin', 0.0024860944)]]

So, as per VGG16 model prediction, the given image may be a tabby (71%) or a tiger cat (21%). You can try the same with different images from ImageNet database and check your results.

eBook - Deep Learning Objective Type Questions and Answers

This book contains 205 objective type questions and answers covering various basic concepts of deep learning. It contains 19 chapters. Each chapter contains a short description of a concept and objective type questions from that concept. 

You can download this book from here.

Please download this book, study and distribute among your friends and colleagues. I would be more than happy if this book can increase your deep learning knowledge to some extent.

Assumption: This should not be your first book on deep learning as I have not covered deep learning concepts in detail, just given a short description to revise your concepts. So, I assume, you have some basic understanding of deep learning concepts before reading this book.

I am continuously upgrading this book and adding more and more objective type deep learning questions in this book as well as in deep learning quiz. So, stay tuned! You can visit this page once in a week and download the updated copy of this book.

Your opinion about this book matters a lot to me. Please post your comments and suggestions regarding this eBook on this blog post.

Table of Contents

Deep Learning Concepts

This book contains objective questions on following Deep Learning concepts:

1. Perceptrons: Working of a Perceptron, multi-layer Perceptron, advantages and limitations of Perceptrons, implementing logic gates like AND, OR and XOR with Perceptrons etc.

2. Neural Networks: Layers in a neural network, types of neural networks, deep and shallow neural networks, forward and backward propagation in a neural network etc.

3. Weights and Bias: Importance of weights and biases, things to keep in mind while initializing weights and biases, Xavier Weight Initialization technique etc.

4. Activation Functions: Importance of activation functions, Squashing functions, Step (Threshold), Logistic (Sigmoid), Hyperbolic Tangent (Tanh), ReLU (Rectified Linear Unit), Dying and Leaky ReLU, Softmax etc.

5. Batches: Epochs, Batches and Iterations, Batch Normalization etc.

6. Gradient Descent: Batch, Stochastic and Mini Batch Gradient Descent, SGD variants like Momentum, Nesterov Momentum, AdaGrad, AdaDelta, RMSprop and Adam, Local and Global Minima, Vanishing and Exploding Gradients, Learning Rate etc.

7. Loss Functions: categorical_crossentropy, sparse_categorical_crossentropy etc.

8. CNN: Convolutional Neural Network, Filters (Kernels), Stride, Padding, Zero Padding and Valid Padding, Pooling, Max Pooling, Min Pooling, Average Pooling and Sum Pooling, Hyperparameters in CNN, Capsule Neural Network (CapsNets), ConvNets vs CapsNets, Computer vision etc.

9. RNN: Recurrent Neural Network, Feedback loop, Types of RNN like One to One, One to Many, Many to One and Many to Many, Bidirectional RNN, Advantages and disadvantages of RNN, Applications of RNN, Differences between CNN and RNN etc.

10. LSTM: Long Short Term Memory, Gated cells like Forget gate, Input gate and Output gate, Applications of LSTM etc.

11. Regularization: Overfitting and underfitting in a neural network, L1 and L2 Regularization, Dropout, Data Augmentation, Early Stopping etc.

12. Fine-tuning: Transfer Learning, Fine-tuning a model, Steps to fine-tune a model, Advantages of fine-tuning etc.

13. Autoencoders: Components of an autoencoder like encoder, decoder and bottleneck, Latent space representation and reconstruction loss, Types of Autoencoders like Undercomplete autoencoder, Sparse autoencoder, Denoising autoencoder, Convolutional autoencoder, Contractive autoencoders and Deep autoencoders, Hyperparameters in an autoencoder, Applications of an Autoencoder, Autoencoders vs PCA, RBM (Restricted Boltzman Machine) etc.

14. NLP (Natural Language Processing): Tokenization, Stemming, Lemmatization and Vectorization (Count vectorization, N-grams vectorization, Term Frequency - Inverse Document Frequency (TF-IDF)), Document-term matrix, NLTK( Natural Language Toolkit) etc.

15. Frameworks: TesnorFlow, Keras, PyTorch, Theano, CNTK, Caffe, MXNet, DL4J etc.

A note to readers

This book is just a short summary of my online contents on:

1. The Professionals Point
2. Online ML Quiz

Contents of this book are available on my this blog (The Professionals Point) and objective type questions are available in the form of quiz on my website (Online ML Quiz). 

Disclaimer: Contents of this book are the sole property of www.onlinemlquiz.com. Questions should not be reproduced in any form without prior permission and attribution.

Solving a regression problem using a Sequential Neural Network Model in Keras

Lets solve a regression problem using neural networks. We will build a sequential model in Keras to predict house prices based on some parameters. We will use KerasRegressor to build a regression model.

You can download housing_data.csv from here. You can also download my Jupyter notebook containing below code of Neural Network Regression implementation.

Step 1: Import required libraries like pandas, numpy, sklearn, keras and matplotlib

import numpy as np
import pandas as pd

from sklearn.preprocessing import MinMaxScaler
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_absolute_error, mean_squared_error

from keras.models import Sequential
from keras.layers import Dense
from keras.wrappers.scikit_learn import KerasRegressor

import matplotlib.pyplot as plt
%matplotlib inline

Step 2: Load and examine the dataset

dataset = pd.read_csv('housing_data.csv')
dataset.head()
dataset.shape
dataset.describe(include='all')

Please note that "describe()" is used to display the statistical values of the data like mean and standard deviation.

Step 3: Mention X and Y axis

X=dataset.iloc[:,0:13]
y=dataset.iloc[:,13].values

X contains the list of attributes
Y contains the list of labels

Step 4: Split the dataset into training and testing dataset

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.20, random_state=0) 

Step 5: Scale the features

scaler = MinMaxScaler()
X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)
y = y.reshape(-1,1)
y = scaler.fit_transform(y)

This step is must for neural networks. Feature scaling is very important for neural networks to perform better and predict accurate results. We should scale both X and y data.

Step 6: Build a neural network

def build_regression_model():
    model = Sequential()
    model.add(Dense(50, input_dim=13, activation='relu'))
    model.add(Dense(50, activation='relu'))
    model.add(Dense(50, activation='relu'))
    model.add(Dense(1, activation='linear'))
    model.compile(optimizer='adam', loss='mean_squared_error')
    return model

We are creating a sequential model with fully connected layers. We are using four layers (one input layer, one output layer and two hidden layers). Input layer and hidden layers are using "relu" activation function while output layer is using "linear" activation function. 

Input layer and hidden layers contain 50 neurons and output layer contains only one neuron as we need to output only one value (predicted house price). You can change the number of neurons in the input and hidden layers as per your data and model performance. Number of hidden layers and number of neurons in each layer are the hyperparameters which you need to tune as the the performance of the model. 

We are using "adam" optimizer and mean square error as a loss function.

We can also use dropout in hidden layers for regularization. But, for this example, I am skipping this step for simplification.

Step 7: Train the neural network

regressor = KerasRegressor(build_fn=build_regression_model, batch_size=32, epochs=150) 
training_history = regressor.fit(X_train,y_train)

We are using 150 epochs with batch size of 32. Number of epochs and batch size are also the hyperparameters which need to be tuned. 

Step 8: Print a loss plot

plt.plot(training_history.history['loss'])
plt.show()

This plot shows that after around 140 epochs, the loss does not vary so much. That is why, I have taken number of epochs as 150 in step 7 while training the neural network.

Step 9: Predict from the neural network

y_pred= regressor.predict(X_test)
y_pred

The y_pred is a numpy array that contains all the predicted values for the input values in the X_test.

Lets see the difference between the actual and predicted values.

df=pd.DataFrame({'Actual':y_test, 'Predicted':y_pred})  
df 

Step 10: Check the accuracy

meanAbsoluteError = mean_absolute_error(y_test, y_pred)
meanSquaredError = mean_squared_error(y_test, y_pred)
rootMeanSquaredError = np.sqrt(meanSquaredError)
print('Mean Absolute Error:', meanAbsoluteError)  
print('Mean Squared Error:', meanSquaredError)  
print('Root Mean Squared Error:', rootMeanSquaredError)

Output:
Mean Absolute Error: 2.9524098807690193 Mean Squared Error: 19.836363961675836 Root Mean Squared Error: 4.453803314210882

We have got the root mean square error as 4.45. We can further decrease this error using cross validation and tuning our hyperparameters. I am leaving it for you to practice.

Step 11: Visualize the results using scatter plot 

plt.scatter(range(len(y_test)), y_test, c='g')
plt.scatter(range(len(y_test)), y_pred, c='b')
plt.xlabel('Test data')
plt.ylabel('Predicted data')
plt.show()

We are displaying test labels and predicted values in different colors (green and blue). From the scatter plot, we can visualize that our neural network has done a great job.

Step 12: Visualize results using regression plot

To further visualize the predicted results, we can draw a regression plot.

fig, ax = plt.subplots()
ax.scatter(y_test, y_pred)
ax.plot([y_test.min(), y_test.max()], [y_test.min(), y_test.max()], 'k--', lw=4)
ax.set_xlabel('Test data')
ax.set_ylabel('Predicted data')
plt.show()

I hope, I was able to demonstrate this regression problem to a large extent. If you have further any doubt, please post a comment.