#Build a Basic ML Model for Text Classification #In this notebook, you'll learn how to implement a text classification task using machine learning. #You'll learn to create basic NLP based features that can be created from the text and you'll then test the model on the test data set to evaluate it's performance. #To make things interesting, the task is to build a machine learning model to classify whether a particular tweet is hate speech or not. I'll explain more as you proceed further, so let's start without much ado! #Table of Contents #About the Dataset #Text Cleaning #Feature Engineering #Train an ML model for Text Classification #Evaluate the ML model #Conclusion #1. About the Dataset #The dataset that you are going to use is of Detecting Hate Speech in people's tweets. #You can download it from here. Let's load the dataset using pandas and have a quick look at some sample tweets. #Load the dataset import pandas as pd
dataset = pd.read_csv('final_dataset_basicmlmodel.csv') dataset.head() #Things to note
#label is the column that contains the target variable or the value that has to be predicted. #1 means it's a hate speech and 0 means it is not. #tweet is the column that contains the text of the tweet. #This is the main data on which NLP techniques will be applied. #Let's have a close look at some of the tweets. for index, tweet in enumerate(dataset["tweet"][10:15]): print(index+1,".",tweet) #Note :- Noise present in Tweets
#If you look closely, you'll see that there are many hashtags present in the tweets of the form # symbol followed by text. #We particularly don't need the # symbol so we will clean it out. #Also, there are strange symbols like â and ð in tweet 4. #This is actually unicode characters that is present in our dataset that we need to get rid of #because they don't particularly add anything meaningful. #There are also numerals and percentages . #2. Data Cleaning #Let's clean up the noise in our dataset. import re
#Clean text from noise def clean_text(text): #Filter to allow only alphabets text = re.sub(r'[^a-zA-Z\']', ' ', text)
#Remove Unicode characters
text = re.sub(r'[^\x00-\x7F]+', '', text)
#Convert to lowercase to maintain consistency
text = text.lower()
return text
dataset['clean_text'] = dataset.tweet.apply(lambda x: clean_text(x)) #3. Feature Engineering #Feature engineering is the science (and art) of extracting more information from existing data. #You are not adding any new data here, but you are actually making the data you already have more useful. #The machine learning model does not understand text directly, so we create numerical features that reperesant the underlying text. #In this module, you'll deal with very basic NLP based features #and as you progress further in the course you'll come across more complex and efficient ways of doing the same. #Exhaustive list of stopwords in the english language. We want to focus less on these so at some point will have to filter STOP_WORDS = ['a', 'about', 'above', 'after', 'again', 'against', 'all', 'also', 'am', 'an', 'and', 'any', 'are', "aren't", 'as', 'at', 'be', 'because', 'been', 'before', 'being', 'below', 'between', 'both', 'but', 'by', 'can', "can't", 'cannot', 'com', 'could', "couldn't", 'did', "didn't", 'do', 'does', "doesn't", 'doing', "don't", 'down', 'during', 'each', 'else', 'ever', 'few', 'for', 'from', 'further', 'get', 'had', "hadn't", 'has', "hasn't", 'have', "haven't", 'having', 'he', "he'd", "he'll", "he's", 'her', 'here', "here's", 'hers', 'herself', 'him', 'himself', 'his', 'how', "how's", 'however', 'http', 'i', "i'd", "i'll", "i'm", "i've", 'if', 'in', 'into', 'is', "isn't", 'it', "it's", 'its', 'itself', 'just', 'k', "let's", 'like', 'me', 'more', 'most', "mustn't", 'my', 'myself', 'no', 'nor', 'not', 'of', 'off', 'on', 'once', 'only', 'or', 'other', 'otherwise', 'ought', 'our', 'ours', 'ourselves', 'out', 'over', 'own', 'r', 'same', 'shall', "shan't", 'she', "she'd", "she'll", "she's", 'should', "shouldn't", 'since', 'so', 'some', 'such', 'than', 'that', "that's", 'the', 'their', 'theirs', 'them', 'themselves', 'then', 'there', "there's", 'these', 'they', "they'd", "they'll", "they're", "they've", 'this', 'those', 'through', 'to', 'too', 'under', 'until', 'up', 'very', 'was', "wasn't", 'we', "we'd", "we'll", "we're", "we've", 'were', "weren't", 'what', "what's", 'when', "when's", 'where', "where's", 'which', 'while', 'who', "who's", 'whom', 'why', "why's", 'with', "won't", 'would', "wouldn't", 'www', 'you', "you'd", "you'll", "you're", "you've", 'your', 'yours', 'yourself', 'yourselves']
#Generate word frequency def gen_freq(text): #Will store the list of words word_list = []
#Loop over all the tweets and extract words into word_list
for tw_words in text.split():
word_list.extend(tw_words)
#Create word frequencies using word_list
word_freq = pd.Series(word_list).value_counts()
#Drop the stopwords during the frequency calculation
word_freq = word_freq.drop(STOP_WORDS, errors='ignore')
return word_freq
#Check whether a negation term is present in the text def any_neg(words): for word in words: if word in ['n', 'no', 'non', 'not'] or re.search(r"\wn't", word): return 1 else: return 0
#Check whether one of the 100 rare words is present in the text def any_rare(words, rare_100): for word in words: if word in rare_100: return 1 else: return 0
#Check whether prompt words are present def is_question(words): for word in words: if word in ['when', 'what', 'how', 'why', 'who']: return 1 else: return 0
word_freq = gen_freq(dataset.clean_text.str) #100 most rare words in the dataset rare_100 = word_freq[-100:] #Number of words in a tweet dataset['word_count'] = dataset.clean_text.str.split().apply(lambda x: len(x)) #Negation present or not dataset['any_neg'] = dataset.clean_text.str.split().apply(lambda x: any_neg(x)) #Prompt present or not dataset['is_question'] = dataset.clean_text.str.split().apply(lambda x: is_question(x)) #Any of the most 100 rare words present or not dataset['any_rare'] = dataset.clean_text.str.split().apply(lambda x: any_rare(x, rare_100)) #Character count of the tweet dataset['char_count'] = dataset.clean_text.apply(lambda x: len(x)) #Top 10 common words are gen_freq(dataset.clean_text.str)[:10] dataset.head() #Splitting the dataset into Train-Test split #The dataset is split into train and test sets so that we can evaluate our model's performance on unseen data. #The model will only be trained on the train set and will make predictions on the test set whose data points the model has never seen. #This will make sure that we have a proper way to test the model. #This is a pretty regular practice in Machine Learning, don't worry if you are confused. #It's just a way of testing your model's performance on unseen data. from sklearn.model_selection import train_test_split
X = dataset[['word_count', 'any_neg', 'any_rare', 'char_count', 'is_question']] y = dataset.label
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=27) #4. Train an ML model for Text Classification #Now that the dataset is ready, it is time to train a Machine Learning model on the same. #You will be using a Naive Bayes classifier from sklearn which is a prominent python library used for machine learning.
from sklearn.naive_bayes import GaussianNB
#Initialize GaussianNB classifier model = GaussianNB() #Fit the model on the train dataset model = model.fit(X_train, y_train) #Make predictions on the test dataset pred = model.predict(X_test)
#5. Evaluate the ML model #It is time to train the model on previously unseen data: X_test and y_test sets that you previously created. #Let's check the accuracy of the model. from sklearn.metrics import accuracy_score
print("Accuracy:", accuracy_score(y_test, pred)*100, "%")
#6. Conclusion #Note: that since we have used very basic NLP features, #the classification accuracy and f1 scores aren't that impressive. #The goal of this exercise was to make you familiar with the model building process #and I hope that you have a better idea on how to build a text classification model.