Sentiment analysis of book reviews

With the rise in e-commerce, popularity of shopping vendors like Amazon is on rise.
Often at times, customers express their opinion or sentiment by giving feedback or reviews.
These reviews, or feedbacks are in the form of text.
Sentiment analysis is the process of determining the opinion, reviews or feeling expressed as either positive, negative or neutral.
Capturing the exact sentiment of a review through text is a challenging task.
In this notebook, various preprocessing techniques like HTML tags and URLs removal, punctuation, whitespace, special character removal and stemming are used to clean the reviews.
The preprocessed data is represented using feature selection techniques like term frequency-inverse document frequency (TF–IDF).
The classifiers like Decision Tree (DT), Support Vector Machine (SVM), Linear regression (RF) and Naive Bayes (NB) are used to classify sentiment of Amazon book reviews.
Finally, (i) comparison of various classifiers based on F1 Score and Accuracy, (ii) Tune the selected model using grid-search and (iii) Performed classification on unseen data.

Data Class

import random

class Sentiment:
    NEGATIVE = "NEGATIVE"
    NEUTRAL = "NEUTRAL"
    POSITIVE = "POSITIVE"
    
#instead of selectin using indexing from the text and score we can create a class 

class Review:
    def __init__(self, text, score):
        self.text = text
        self.score = score
        self.sentiment = self.get_sentiment()
        
    def get_sentiment(self):
        if self.score <= 2:
            return Sentiment.NEGATIVE
        elif self.score == 3:
            return Sentiment.NEUTRAL
        else: #Score of 4 or 5
            return Sentiment.POSITIVE
        
class ReviewContainer:
    def __init__(self, reviews):
        self.reviews = reviews
        
    def get_text(self):
        return [x.text for x in self.reviews]
    
    def get_sentiment(self):
        return [x.sentiment for x in self.reviews]
        
    def evenly_distribute(self):
        negative = list(filter(lambda x: x.sentiment == Sentiment.NEGATIVE, self.reviews))
        positive = list(filter(lambda x: x.sentiment == Sentiment.POSITIVE, self.reviews))
        positive_shrunk = positive[:len(negative)]
        self.reviews = negative + positive_shrunk
        random.shuffle(self.reviews)
        

Load data

import json

file_name = 'Books_small_10000.json'

with open(file_name) as f:
    for line in f:
        review = json.loads(line)
        print(review['reviewText'])
        print(review['overall'])
        break

I bought both boxed sets, books 1-5.  Really a great series!  Start book 1 three weeks ago and just finished book 5.  Sloane Monroe is a great character and being able to follow her through both private life and her PI life gets a reader very involved!  Although clues may be right in front of the reader, there are twists and turns that keep one guessing until the last page!  These are books you won't be disappointed with.
5.0

reviews = []
with open(file_name) as f:
    for line in f:
        review = json.loads(line)
        reviews.append(Review(review['reviewText'], review['overall']))
        
reviews[5].sentiment

'POSITIVE'

Prep Data

from sklearn.model_selection import train_test_split

training, test = train_test_split(reviews, test_size=0.33, random_state=42)

train_container = ReviewContainer(training)

test_container = ReviewContainer(test)

train_container.evenly_distribute()
train_x = train_container.get_text()
train_y = train_container.get_sentiment()

test_container.evenly_distribute()
test_x = test_container.get_text()
test_y = test_container.get_sentiment()

print(train_y.count(Sentiment.POSITIVE))
print(train_y.count(Sentiment.NEGATIVE))

436
436

Bag of words vectorization

from sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer

# This book is great !
# This book was so bad

vectorizer = TfidfVectorizer()
train_x_vectors = vectorizer.fit_transform(train_x)

test_x_vectors = vectorizer.transform(test_x)

print(train_x[0])
print(train_x_vectors[0].toarray())

The book isn't in very good condition. The pages are yellowed. The spine is well worn. Not that good a shape.
[[0. 0. 0. ... 0. 0. 0.]]

Classification

Linear SVM

from sklearn import svm

clf_svm = svm.SVC(kernel='linear')

clf_svm.fit(train_x_vectors, train_y)

test_x[0]

clf_svm.predict(test_x_vectors[0])

array(['NEGATIVE'], dtype='<U8')

Decision Tree Classifier

from sklearn.tree import DecisionTreeClassifier

clf_dec = DecisionTreeClassifier()
clf_dec.fit(train_x_vectors, train_y)

clf_dec.predict(test_x_vectors[0])

array(['POSITIVE'], dtype='<U8')

naive bayes

from sklearn.naive_bayes import GaussianNB

clf_gnb = DecisionTreeClassifier()
clf_gnb.fit(train_x_vectors, train_y)

clf_gnb.predict(test_x_vectors[0])

array(['POSITIVE'], dtype='<U8')

Logistic Regression

from sklearn.linear_model import LogisticRegression

clf_log = LogisticRegression()
clf_log.fit(train_x_vectors, train_y)

clf_log.predict(test_x_vectors[0])

array(['NEGATIVE'], dtype='<U8')

Evluation

# Mean Accuracy
print(clf_svm.score(test_x_vectors, test_y))
print(clf_dec.score(test_x_vectors, test_y))
print(clf_gnb.score(test_x_vectors, test_y))
print(clf_log.score(test_x_vectors, test_y))

0.8076923076923077
0.6298076923076923
0.6370192307692307
0.8052884615384616

# F1 Scores
from sklearn.metrics import f1_score

f1_score(test_y, clf_svm.predict(test_x_vectors), average=None, labels=[Sentiment.POSITIVE, Sentiment.NEGATIVE])
#f1_score(test_y, clf_log.predict(test_x_vectors), average=None, labels=[Sentiment.POSITIVE, Sentiment.NEUTRAL, Sentiment.NEGATIVE])

array([0.80582524, 0.80952381])

test_set = ['very fun', "bad book do not buy", 'horrible waste of time']
new_test = vectorizer.transform(test_set)

clf_svm.predict(new_test)

array(['POSITIVE', 'NEGATIVE', 'NEGATIVE'], dtype='<U8')

Tuning our model (with Grid Search)

from sklearn.model_selection import GridSearchCV

parameters = {'kernel': ('linear', 'rbf'), 'C': (1,4,8,16,32)}

svc = svm.SVC()
clf = GridSearchCV(svc, parameters, cv=5)
clf.fit(train_x_vectors, train_y)

GridSearchCV(cv=5, estimator=SVC(),
             param_grid={'C': (1, 4, 8, 16, 32), 'kernel': ('linear', 'rbf')})

print(clf.score(test_x_vectors, test_y))

0.8100961538461539

Saving Model

import pickle

with open('sentiment_classifier.pkl', 'wb') as f:
    pickle.dump(clf, f)

Load model

with open('sentiment_classifier.pkl', 'rb') as f:
    loaded_clf = pickle.load(f)

print(test_x[1])

loaded_clf.predict(test_x_vectors[1])

Some of the characters seem like an amalgamation of familiar video game characters and stereotypes but over Id like to live in this universe. Very good series.

array(['POSITIVE'], dtype='<U8')

Raw Data download:

http://jmcauley.ucsd.edu/data/amazon/

Description

This dataset contains product reviews and metadata from Amazon, including 142.8 million reviews spanning May 1996 - July 2014.
This dataset includes reviews (ratings, text, helpfulness votes), product metadata (descriptions, category information, price, brand, and image features), and links (also viewed/also bought graphs).
References1!..