Sentiment Analysis

of Product-Based Reviews
Using Machine Learning
Approaches
BY
ANUSUYA DHARA (CSE/2014/041)
ARKADEB SAHA (CSE/2014/048)
SOURISH SENGUPTA (CSE/2014/049)
PRANIT BOSE (CSE/2014/060)

UNDER THE GUIDANCE OF

Dr. Anup Kumar Kolya

PROJECT REPORT SUBMITTED IN PARTIAL FULFILLMENT OF

THE REQUIREMENTS FOR THE DEGREE OF
BACHELOR OF TECHNOLOGY IN COMPUTER SCIENCE AND ENGINEERING
RCC INSTITUTE OF INFORMATION TECHNOLOGY
Session 2017-2018

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

RCC INSTITUTE OF INFORMATION TECHNOLOGY

[Affiliated to West Bengal University of Technology]
CANAL SOUTH ROAD, BELIAGHATA, KOLKATA-700015
 DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING
RCC INSTITUTE OF INFORMATION TECHNOLOGY

TO WHOM IT MAY CONCERN

I hereby recommend that the project Sentimental Analysis of Product-Based Reviews

using Machine Learning Approaches prepared under my supervision by Anusuya Dhara
(University Roll No.:11700114008 | Class Roll No.: CSE/2014/041), Arkadeb Saha
(University Roll No.:11700114012 | Class Roll No.: CSE/2014/048), Sourish Sengupta
(University Roll No.:11700114073 | Class Roll No.: CSE/2014/049) and Pranit Bose
(University Roll No.:11700114044 | Class Roll No.: CSE/2014/060) of B.Tech (8th
Semester), may be accepted in partial fulfillment for the degree of Bachelor of Technology
in Computer Science and Engineering under Maulana Abul Kalam Azad University of
Technology (MAKAUT).

………………………………………
Project Supervisor
Department of Computer Science and Engineering
RCC Institute of Information Technology
Countersigned:

………………………………………
Head
Department of Computer Science & Engineering
RCC Institute of Information Technology
Kolkata – 700015

2
 DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING
RCC INSTITUTE OF INFORMATION TECHNOLOGY

CERTIFICATE OF APPROVAL

The foregoing Project is hereby accepted as a credible study of an engineering

subject carried out and presented in a manner satisfactory to warrant its acceptance as a
prerequisite to the degree for which it has been submitted. It is understood that by this
approval the undersigned do not necessarily endorse or approve any statement made,
opinion expressed or conclusion drawn therein, but approve the project only for the purpose
for which it is submitted.

FINAL EXAMINATION FOR 1. —————————————

EVALUATION OF PROJECT

2. ———————————

(Signature of Examiners)

3
 ACKNOWLEDGEMENT

We would like to express special thanks & gratitude to our guide, Dr. Anup Kumar Kolya
who gave us this golden opportunity to work on this scalable project on the topic of
“Sentiment Analysis of product based reviews using Machine Learning Approaches”, which
led us into doing a lot of Research which diversified our knowledge to a huge extent for
which we are thankful.

Also, we would like to thank our parents and friends who supported us a lot in finalizing
this project within the limited time frame.

--------------------------------------------------
ANUSUYA DHARA (CSE/2014/041)

--------------------------------------------------
ARKADEB SAHA (CSE/2014/048)

--------------------------------------------------
SOURISH SENGUPTA (CSE/2014/049)

--------------------------------------------------
PRANIT BOSE (CSE/2014/060)

4
 Table of Contents

Page No.

1. Abstract …………………………………………………………………… 6

2. Introduction ………………………………………………………………. 7

3. Review of Literature ……………………………………………………… 8

4. Objective of the Project …………………………………………………... 9

5. System Design……………………………………………………………… 10

6. Methodology for implementation (Formulation/Algorithm)………….... 12

7. Implementation Details ……………………………………………………. 17

8. Results/Sample output ……………………………………………………... 20

9. Conclusion…………………………………………………………………… 22

Appendix-: Program Source code with adequate comments. 23

References 27

5
 1. Abstract

Sentiment Analysis also known as Opinion Mining refers to the use of natural language
processing, text analysis to systematically identify, extract, quantify, and study affective
states and subjective information. Sentiment analysis is widely applied to reviews and
survey responses, online and social media, and healthcare materials for applications that
range from marketing to customer service to clinical medicine.

In this project, we aim to perform Sentiment Analysis of product based reviews. Data used
in this project are online product reviews collected from “amazon.com”. We expect to do
review-level categorization of review data with promising outcomes.

6
 1. Introduction

Sentiment is an attitude, thought, or judgment prompted by feeling.

Sentiment analysis, which is also known as opinion mining, studies people’s
sentiments towards certain entities. From a user’s perspective, people are able to post
their own content through various social media, such as forums, micro-blogs, or
online social networking sites. From a researcher’s perspective, many social media
sites release their application programming interfaces (APIs), prompting data
collection and analysis by researchers and developers. However, those types of
online data have several flaws that potentially hinder the process of sentiment
analysis. The first flaw is that since people can freely post their own content, the
quality of their opinions cannot be guaranteed. he second flaw is that ground truth of
such online data is not always available. A ground truth is more like a tag of a certain
opinion, indicating whether the opinion is positive, negative, or neutral.

“It is a quite boring movie…….. but the scenes were good enough. ”
The given line is a movie review that states that “it” (the movie) is
quite boring but the scenes were good. Understanding such sentiments require
multiple tasks.
Hence, SENTIMENTAL ANALYSIS is a kind of text classification
based on Sentimental Orientation (SO) of opinion they contain.
Sentiment analysis of product reviews has recently become very popular in text
mining and computational linguistics research.

 Firstly, evaluative terms expressing opinions must be extracted from the

review.

 Secondly, the SO, or the polarity, of the opinions must be determined.

 Thirdly, the opinion strength, or the intensity, of an opinion should also be

determined.

 Finally, the review is classified with respect to sentiment classes, such as

Positive and Negative, based on the SO of the opinions it contains.

7
 2. Review of Literature

The most fundamental problem in sentiment analysis is the sentiment polarity

categorization, by considering a dataset containing over 5.1 million product reviews
from Amazon.com with the products belonging to four categories.

A max-entropy POS tagger is used in order to classify the words of the sentence, an
additional python program to speed up the process. The negation words like no, not,
and more are included in the adverbs whereas Negation of Adjective and Negation of
Verb are specially used to identify the phrases.

The following are the various classification models which are selected for
categorization: Naïve Bayesian, Random Forest, Logistic Regression and Support
Vector Machine.

For feature selection, Pang and Lee suggested to remove objective sentences by
extracting subjective ones. They proposed a text-categorization technique that is able
to identify subjective content using minimum cut. Gann et al. selected 6,799 tokens
based on Twitter data, where each token is assigned a sentiment score, namely TSI
(Total Sentiment Index), featuring itself as a positive token or a negative token.
Specifically, a TSI for a certain token is computed as:

where p is the number of times a token appears in positive tweets and n is the
number of times a token appears in negative tweets is the ratio of total

number of positive tweets over total number of negative tweets.

8
 3. Objective of the Project

Scrapping product reviews on various websites featuring various products

specifically amazon.com.
Analyze and categorize review data.

Analyze sentiment on dataset from document level (review level).

Categorization or classification of opinion sentiment into-

 Positive
 Negative

9
 4. System Design

Hardware Requirements:
 Core i5/i7 processor
 At least 8 GB RAM
 At least 60 GB of Usable Hard Disk Space

Software Requirements:
 Python 3.x
 Anaconda Distribution
 NLTK Toolkit
 UNIX/LINUX Operating System.

Data Information:

 The Amazon reviews dataset consists of reviews from amazon. The data span
a period of 18 years, including ~35 million reviews up to March 2013.
Reviews include product and user information, ratings, and a plaintext review.
For more information, please refer to the following paper: J. McAuley and J.
Leskovec. Hidden factors and hidden topics: understanding rating dimensions
with review text. RecSys, 2013.

 The Amazon reviews full score dataset is constructed by Xiang Zhang

([email protected]) from the above dataset. It is used as a text
classification benchmark in the following paper: Xiang Zhang, Junbo Zhao,
Yann LeCun. Character-level Convolutional Networks for Text Classification.
Advances in Neural Information Processing Systems 28 (NIPS 2015).

 The Amazon reviews full score dataset is constructed by randomly taking

200,000 samples for each review score from 1 to 5. In total there are
1,000,000 samples.

10
Data Format:
The dataset we will use is .json file. The sample of the dataset is given below.
{
"reviewSummary": "Surprisingly delightful",
"reviewText": “ This is a first read filled with unexpected humor and profound
insights into the art of politics and policy. In brief, it is sly, wry, and wise. ”,
"reviewRating": “4”,

11
 5. Methodology for Implementation
(Formulation/Algorithm)

DATA COLLECTION:
Data which means product reviews collected from amazon.com from May
1996 to July 2014. Each review includes the following information: 1) reviewer ID; 2)
product ID; 3) rating; 4) time of the review; 5) helpfulness; 6) review text. Every rating is
based on a 5-star scale, resulting all the ratings to be ranged from 1-star to 5-star with no
existence of a half-star or a quarter-star.

SENTIMENT SENTENCE EXTRACTION & POS TAGGING:

Tokenization of reviews after removal of STOP words which mean nothing
related to sentiment is the basic requirement for POS tagging. After proper removal of
STOP words like “am, is, are, the, but” and so on the remaining sentences are converted in
tokens. These tokens take part in POS tagging
In natural language processing, part-of-speech (POS) taggers have been
developed to classify words based on their parts of speech. For sentiment analysis, a POS
tagger is very useful because of the following two reasons: 1) Words like nouns and
pronouns usually do not contain any sentiment. It is able to filter out such words with the
help of a POS tagger; 2) A POS tagger can also be used to distinguish words that can be
used in different parts of speech.

NEGETIVE PHRASE IDENTIFICATION:

Words such as adjectives and verbs are able to convey opposite sentiment
with the help of negative prefixes. For instance, consider the following sentence that was
found in an electronic device’s review: “The built in speaker also has its uses but so far
nothing revolutionary." The word, “revolutionary" is a positive word according to the list in.
However, the phrase “nothing revolutionary" gives more or less negative feelings.
Therefore, it is crucial to identify such phrases. In this work, there are two types of phrases
have been identified, namely negation-of-adjective (NOA) and negation-of-verb (NOV).

12
SENTIMENT CLASSIFICATION ALGORITHMS:

Naïve Bayesian classifier:

The Naïve Bayesian classifier works as follows: Suppose that there exist a set of
training data, D, in which each tuple is represented by an n-dimensional feature
vector, X=x 1,x 2,..,x n , indicating n measurements made on the tuple from n attributes or
features. Assume that there are m classes, C 1,C 2,...,C m . Given a tuple X, the classifier will
predict that X belongs to C i if and only if: P(C i |X)>P(C j |X),
where i,j∈[1,m]a n d i≠j. P(C i |X) is computed as:

Random forest

The random forest classifier was chosen due to its superior performance over a single
decision tree with respect to accuracy. It is essentially an ensemble method based on
bagging. The classifier works as follows: Given D, the classifier firstly creates k bootstrap
samples of D, with each of the samples denoting as Di . A Di has the same number of tuples
as D that are sampled with replacement from D. By sampling with replacement, it means
that some of the original tuples of D may not be included in Di , whereas others may occur
more than once. The classifier then constructs a decision tree based on each Di . As a result,

13
a “forest" that consists of k decision trees is formed.

To classify an unknown tuple, X, each tree returns its class prediction counting as one vote.
The final decision of X’s class is assigned to the one that has the most votes.

The decision tree algorithm implemented in scikit-learn is CART (Classification and

Regression Trees). CART uses Gini index for its tree induction. For D, the Gini index is
computed as:

Where pi is the probability that a tuple in D belongs to class C i . The Gini index measures
the impurity of D. The lower the index value is, the better D was partitioned.

Support vector machine

Support vector machine (SVM) is a method for the classification of both linear and
nonlinear data. If the data is linearly separable, the SVM searches for the linear optimal
separating hyperplane (the linear kernel), which is a decision boundary that separates data
of one class from another. Mathematically, a separating hyper plane can be written
as: W·X+b=0, where W is a weight vector and W=w1,w2,...,w n. X is a training tuple. b is a
scalar. In order to optimize the hyperplane, the problem essentially transforms to the
minimization of ∥W∥, which is eventually computed as:

where αi are numeric parameters, and yi are labels based on support

vectors, Xi .
That is: if yi =1 then

if y i =−1 then

14
If the data is linearly inseparable, the SVM uses nonlinear mapping to transform the data
into a higher dimension. It then solve the problem by finding a linear hyperplane. Functions
to perform such transformations are called kernel functions. The kernel function selected for
our experiment is the Gaussian Radial Basis Function (RBF):

where Xi are support vectors, X j are testing tuples, and γ is a free parameter that uses the
default value from scikit-learn in our experiment. Figure shows a classification example of
SVM based on the linear kernel and the RBF kernel on the next page-

Logistic Regression

Logistic regression predicts the probability of an outcome that can only have two
values (i.e. a dichotomy). The prediction is based on the use of one or several predictors
(numerical and categorical). A linear regression is not appropriate for predicting the
value of a binary variable for two reasons:

15
  A linear regression will predict values outside the acceptable range (e.g. predicting
probabilities outside the range 0 to 1)
 Since the dichotomous experiments can only have one of two possible values
for each experiment, the residuals will not be normally distributed about
the predicted line.
On the other hand, a logistic regression produces a logistic curve, which is limited to
values between 0 and 1. Logistic regression is similar to a linear regression, but the
curve is constructed using the natural logarithm of the “odds” of the target variable,
rather than the probability. Moreover, the predictors do not have to be normally
distributed or have equal variance in each group.

Logistic regression uses maximum likelihood estimation (MLE) to obtain the model
coefficients that relate predictors to the target. After this initial function is estimated,
the process is repeated until LL (Log Likelihood) does not change significantly.

16
 6. Implementation Details
The training of dataset consists of the following steps:

Unpacking of data: The huge dataset of reviews obtained from amazon.com

comes in a .json file format. A small python code has been implemented in order to
read the dataset from those files and dump them in to a pickle file for easier and
fastaccess and object serialization.

Hence initial fetching of data is done in this section using Python File Handlers.

Preparing Data for Sentiment Analysis:

i) The pickle file is hence loaded in this step and the data besides the one used
for sentiment analysis is removed. As shown in our sample dataset in Page 11, there
are a lot of columns in the data out of which only rating and text review is what we
require. So, the column, “reviewSummary” is dropped from the data file.

ii) After that, the review ratings which are 3 out of 5 are removed as they
signify neutral review, and all we are concerned of is positive and negative reviews.
iii) The entire task of preprocessing the review data is handled by this

17
utility class- “NltkPreprocessor”.

iv) The time required to prepare the following data is hence displayed.

The time taken to preprocess the data is calculated and displayed

Preprocessing Data:This is a vital part of training the dataset. Here Words present
in the file are accessed both as a solo word and also as pair of words. Because, for
example the word “bad” means negative but when someone writes “not bad” it refers
to as positive. In such cases considering single word for training data will work
otherwise. So words in pairs are checked to find the occurrence to modifiers before

18
 any adjective which if present which might provide a different meaning to the
outlook.

Training Data/ Evaluation:The main chunk of code that does the whole
evaluation of sentimental analysis based on the preprocessed data is a part of this.
The following are the steps followed:

i) The Accuracy, Precision, Recall, and Evaluation time is calculated and displayed.

ii) Navie Bayes, Logistic Regression, Linear SVM and Random forest classifiers are
applied on the dataset for evaluation of sentiments.

iii) Prediction of test data is done and Confusion Matrix of prediction isdisplayed.
iv) Total positive and negative reviews are counted.
v) A review like sentence is taken as input on the console and if positive the console gives 1
as output and 0 for negative input.

19
 7. Results and Sample Output
The ultimate outcome of this Training of Public reviews dataset is that, the
machine is capable of judging whether an entered sentence bears positive response or
negative response.
Precision (also called positive predictive value) is the fraction of relevant
instances among the retrieved instances, while Recall (also known as sensitivity) is
the fraction of relevant instances that have been retrieved over the total amount of
relevant instances. Both precision and recall are therefore based on an understanding
and measure of relevance.

F1 score (also F-score or F-measure) is a measure of a test's accuracy. It

considers both the precision p and the recall r of the test to compute the score: p is
the number of correct positive results divided by the number of all positive results
returned by the classifier, and r is the number of correct positive results divided by
the number of all relevant samples (all samples that should have been identified as
positive). The F1 score is the harmonic average of the precision and recall, where an
F1 score reaches its best value at 1 (perfect precision and recall) and worst at 0.

In statistics, a receiver operating characteristic curve, i.e. ROC curve, is a

graphical plot that illustrates the diagnostic ability of a binary classifier system as its
discrimination threshold is varied. The Total Operating Characteristic (TOC)
expands on the idea of ROC by showing the total information in the two-by-two
contingency table for each threshold. ROC gives only two bits of relative
information for each threshold, thus the TOC gives strictly more information than the
ROC.

20
 When using normalized units, the area under the curve (often referred to as
simply the AUC) is equal to the probability that a classifier will rank a randomly
chosen positive instance higher than a randomly chosen negative one (assuming
'positive' ranks higher than 'negative'). This can be seen as follows: the area under the
curve is given by (the integral boundaries are reversed as large T has a lower value
on the x-axis).

The machine evaluates the accuracy of training the data along with precision
Recall and F1
The Confusion matrix of evaluation is calculated.
It is thus capable of judging an externally written review as positive or
negative.
A positive review will be marked as [1], and a negative review will be hence
marked as [0].

Results obtained using Hold-out Strategy(Train-Test split) [values rounded upto

2 decimal places].

Name of F1 Accuracy Precision Recall ROC AUC

classifier
Multinomial 85.25% 85.31% 85.56% 84.95% 85.31%
NB
Logistic 88.12% 88.05% 87.54% 88.72% 88.05%
Regression
Linear 88.12% 88.11% 87.59% 88.80% 88.11%
SVC
Random 82.43% 81.82% 79.74% 85.30% 81.83%
Forest

The Confusion Matrix Format is as follows: True

Negative False Positive

False
True Positive
Negative

21
The Confusion Matrix of Each Classifier are as follows:

68556 69928
11470 10098

12032 67942 9023 70951

Classifier 1: Multinomial NB Classifier 2: Logistic Regression

69963 62695
10063 17331

8955 17019 11749 68225

Classifier 3: Liner SVC Classifier 4: Random Forest

22
The following are the images of such sample output after successful dataset training
using the classifiers:

23
24
The Bar Graph showing the Frequency of Ratings in the dataset

This Bar graph shows the score of each classifier after successful training. The parameters
be: F1 Score, Accuracy, Precision, Recall and Roc-Auc.

25
 8. Conclusion

Sentiment analysis deals with the classification of texts based on the sentiments they
contain. This article focuses on a typical sentiment analysis model consisting of three
core steps, namely data preparation, review analysis and sentiment classification, and
describes representative techniques involved in those steps.

Sentiment analysis is an emerging research area in text mining and computational

linguistics, and has attracted considerable research attention in the past few years.
Future research shall explore sophisticated methods for opinion and product feature
extraction, as well as new classification models that can address the ordered labels
property in rating inference. Applications that utilize results from sentiment analysis
is also expected to emerge in the near future.

26
Appendix
Code:
Loading the dataset:
import json
import pickle
import numpy as np
from matplotlib import pyplot as plt
from textblob import TextBlob

# fileHandler = open('datasets/reviews_digital_music.json', 'r')

# reviewDatas = fileHandler.read().split('\n')
# reviewText = []
# reviewRating = []

# for review in reviewDatas:

# if review == "":
# continue
# r = json.loads(review)
# reviewText.append(r['reviewText'])
# reviewRating.append(r['overall'])

# fileHandler.close()
# saveReviewText = open('review_text.pkl', 'wb')
# saveReviewRating = open('review_rating.pkl','wb')
# pickle.dump(reviewText, saveReviewText)
# pickle.dump(reviewRating, saveReviewRating)
reviewTextFile = open('review_text.pkl', 'rb')

27
reviewRatingFile = open('review_rating.pkl', 'rb')
reviewText = pickle.load(reviewTextFile)
reviewRating = pickle.load(reviewRatingFile)
# print(len(reviewText))
# print(reviewText[0])
# print(reviewRating[0])
# ratings = np.array(reviewRating)
plt.hist(ratings, bins=np.arange(ratings.min(), ratings.max()+2)-0.5, rwidth=0.7)
plt.xlabel('Rating', fontsize=14)
plt.ylabel('Frequency', fontsize=14)
plt.title('Histogram of Ratings', fontsize=18)
plt.show()
lang = {}
i=0
for review in reviewText:
tb = TextBlob(review)
l = tb.detect_language()
if l != 'en':
lang.setdefault(l, [])
lang[l].append(i)
print(i, l)
i += 1
print(lang)

Scrapping data:
from selenium import webdriver
from selenium.webdriver.chrome.options import Options
from bs4 import BeautifulSoup
import openpyxl
class Review():
def __init__(self):

28
 self.rating=""
self.info=""
self.review=""
def scrape():
options = Options()
options.add_argument("--headless") # Runs Chrome in headless mode.
options.add_argument('--no-sandbox') # # Bypass OS security model
options.add_argument('start-maximized')
options.add_argument('disable-infobars')
options.add_argument("--disable-extensions")
driver=webdriver.Chrome(executable_path=r'C:\chromedriver\chromedriver.exe')
url='https://www.amazon.com/Moto-PLUS-5th-Generation-Exclusive/product-
reviews/B0785NN142/ref=cm_cr_arp_d_paging_btm_2?ie=UTF8&reviewerType=all_reviews&pageNumb
er=5'
driver.get(url)

soup=BeautifulSoup(driver.page_source,'lxml')
ul=soup.find_all('div',class_='a-section review')
review_list=[]
for d in ul:
a=d.find('div',class_='a-row')
sib=a.findNextSibling()
b=d.find('div',class_='a-row a-spacing-medium review-data')
'''print sib.text'''
new_r=Review()
new_r.rating=a.text
new_r.info=sib.text
new_r.review=b.text

review_list.append(new_r)
driver.quit()
return review_list
def main():

29
 m = scrape()
i=1
for r in m:

book = openpyxl.load_workbook('Sample.xlsx')
sheet = book.get_sheet_by_name('Sample Sheet')
sheet.cell(row=i, column=1).value = r.rating
sheet.cell(row=i, column=1).alignment = openpyxl.styles.Alignment(horizontal='center',
vertical='center', wrap_text=True)
sheet.cell(row=i, column=3).value = r.info
sheet.cell(row=i, column=3).alignment = openpyxl.styles.Alignment(horizontal='center',
vertical='center', wrap_text=True)
sheet.cell(row=i, column=5).value = r.review.encode('utf-8')
sheet.cell(row=i, column=5).alignment = openpyxl.styles.Alignment(horizontal='center',
vertical='center', wrap_text=True)
book.save('Sample.xlsx')
i=i+1
if __name__ == '__main__':
main()

Preprocessing Data:
import string
from nltk.corpus import stopwords as sw
from nltk.corpus import wordnet as wn
from nltk import wordpunct_tokenize
from nltk import sent_tokenize
from nltk import WordNetLemmatizer
from nltk import pos_tag
class NltkPreprocessor:
def __init__(self, stopwords = None, punct = None, lower = True, strip = True):
self.lower = lower
self.strip = strip
self.stopwords = stopwords or set(sw.words('english'))

30
 self.punct = punct or set(string.punctuation)
self.lemmatizer = WordNetLemmatizer()

def tokenize(self, document):

tokenized_doc = []

for sent in sent_tokenize(document):

for token, tag in pos_tag(wordpunct_tokenize(sent)):
token = token.lower() if self.lower else token
token = token.strip() if self.strip else token
token = token.strip('_0123456789') if self.strip else token
# token = re.sub(r'\d+', '', token)

if token in self.stopwords:
continue

if all(char in self.punct for char in token):

continue

lemma = self.lemmatize(token, tag)

tokenized_doc.append(lemma)

return tokenized_doc

def lemmatize(self, token, tag):

tag = {
'N': wn.NOUN,
'V': wn.VERB,
'R': wn.ADV,
'J': wn.ADJ
}.get(tag[0], wn.NOUN)
return self.lemmatizer.lemmatize(token, tag)

31
Sentiment Analysis:

import ast
import numpy as np
import pandas as pd
import re
from nltk.corpus import stopwords
from nltk.stem import SnowballStemmer
from sklearn.model_selection import train_test_split
from sklearn.feature_selection import SelectKBest, chi2, SelectPercentile, f_classif
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.pipeline import Pipeline
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, roc_auc_score,
confusion_matrix
from sklearn.svm import LinearSVC
# from textblob import TextBlob
from time import time

def getInitialData(data_file):
print('Fetching initial data...')
t = time()

i=0
df = {}
with open(data_file, 'r') as file_handler:
for review in file_handler.readlines():
df[i] = ast.literal_eval(review)
i += 1

reviews_df = pd.DataFrame.from_dict(df, orient = 'index')

reviews_df.to_pickle('reviews_digital_music.pickle')

32
 print('Fetching data completed!')
print('Fetching time: ', round(time()-t, 3), 's\n')

# def filterLanguage(text):
# text_blob = TextBlob(text)
# return text_blob.detect_language()

def prepareData(reviews_df):
print('Preparing data...')
t = time()

reviews_df.rename(columns = {"overall" : "reviewRating"}, inplace=True)

reviews_df.drop(columns = ['reviewerID', 'asin', 'reviewerName', 'helpful', 'summary',
'unixReviewTime', 'reviewTime'], inplace = True)

reviews_df = reviews_df[reviews_df.reviewRating != 3.0] # Ignoring 3-star reviews -> neutral

reviews_df = reviews_df.assign(sentiment = np.where(reviews_df['reviewRating'] >= 4.0, 1, 0)) # 1
-> Positive, 0 -> Negative

stemmer = SnowballStemmer('english')
stop_words = stopwords.words('english')

# print(len(reviews_df.reviewText))
# filterLanguage = lambda text: TextBlob(text).detect_language()
# reviews_df = reviews_df[reviews_df['reviewText'].apply(filterLanguage) == 'en']
# print(len(reviews_df.reviewText))

reviews_df = reviews_df.assign(cleaned = reviews_df['reviewText'].apply(lambda text: '

'.join([stemmer.stem(w) for w in re.sub('[^a-z]+|(quot)+', ' ', text.lower()).split() if w not in stop_words])))
reviews_df.to_pickle('reviews_digital_music_preprocessed.pickle')

33
 print('Preparing data completed!')
print('Preparing time: ', round(time()-t, 3), 's\n')

def preprocessData(reviews_df_preprocessed):
print('Preprocessing data...')
t = time()

X = reviews_df_preprocessed.iloc[:, -1].values
y = reviews_df_preprocessed.iloc[:, -2].values

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 42)

print('Preprocessing data completed!')

print('Preprocessing time: ', round(time()-t, 3), 's\n')

return X_train, X_test, y_train, y_test

def evaluate(y_test, prediction):

print('Evaluating results...')
t = time()

print('Accuracy: {}'.format(accuracy_score(y_test, prediction)))

print('Precision: {}'.format(precision_score(y_test, prediction)))
print('Recall: {}'.format(recall_score(y_test, prediction)))
print('f1: {}'.format(f1_score(y_test, prediction)))

print('Results evaluated!')
print('Evaluation time: ', round(time()-t, 3), 's\n')

# getInitialData('datasets/reviews_digital_music.json')
# reviews_df = pd.read_pickle('reviews_digital_music.pickle')

34
# prepareData(reviews_df)
reviews_df_preprocessed = pd.read_pickle('reviews_digital_music_preprocessed.pickle')
# print(reviews_df_preprocessed.isnull().values.sum()) # Check for any null values

X_train, X_test, y_train, y_test = preprocessData(reviews_df_preprocessed)

print('Training data...')
t = time()

pipeline = Pipeline([
('vect', TfidfVectorizer(ngram_range = (1,2), stop_words = 'english',
sublinear_tf = True)),
('chi', SelectKBest(score_func = chi2, k = 50000)),
('clf', LinearSVC(C = 1.0, penalty = 'l1', max_iter = 3000, dual = False,
class_weight = 'balanced'))
])

model = pipeline.fit(X_train, y_train)

print('Training data completed!')

print('Training time: ', round(time()-t, 3), 's\n')

print('Predicting Test data...')

t = time()

prediction = model.predict(X_test)

print('Prediction completed!')
print('Prediction time: ', round(time()-t, 3), 's\n')

evaluate(y_test, prediction)

print('Confusion matrix: {}'.format(confusion_matrix(y_test, prediction)))

35
print()
l = (y_test == 0).sum() + (y_test == 1).sum()
s = y_test.sum()
print('Total number of observations: ' + str(l))
print('Positives in observation: ' + str(s))
print('Negatives in observation: ' + str(l - s))
print('Majority class is: ' + str(s / l * 100) + '%')

Graph Plotting Code:

import numpy as np
import matplotlib.pyplot as plt
from matplotlib.ticker import MaxNLocator
from collections import namedtuple
n_groups = 5
score_MNB = (85.25, 85.31, 85.56, 84.95, 85.31)
score_LR = (88.12, 88.05, 87.54, 88.72, 88.05)
score_LSVC=(88.12, 88.11, 87.59, 88.80, 88.11)
score_RF=(82.43, 81.82, 79.74, 85.30, 81.83)

#n1=(score_MNB[0], score_LR[0], score_LSVC[0], score_RF[0])

#n2=(score_MNB[1], score_LR[1], score_LSVC[1], score_RF[1])
#n3=(score_MNB[2], score_LR[2], score_LSVC[2], score_RF[2])
#n4=(score_MNB[3], score_LR[3], score_LSVC[3], score_RF[3])
#n5=(score_MNB[4], score_LR[4], score_LSVC[4], score_RF[4])
fig, ax = plt.subplots()
index = np.arange(n_groups)
bar_width = 0.1
opacity = 0.7
error_config = {'ecolor': '0.3'}
rects1 = ax.bar(index,score_MNB, bar_width,
alpha=opacity, color='b',

36
 error_kw=error_config,
label='Multinomial Naive Bayes')
z=index + bar_width
rects2 = ax.bar(z, score_LR, bar_width,
alpha=opacity, color='r',
error_kw=error_config,
label='Logistic Regression')
z=z+ bar_width
rects3 = ax.bar(z, score_LSVC, bar_width,
alpha=opacity, color='y',
error_kw=error_config,
label='Linear SVM')
z=z+ bar_width
rects4 = ax.bar(z, score_RF, bar_width,
alpha=opacity, color='g',
error_kw=error_config,
label='Random Forest')
ax.set_xlabel('Score Parameters')
ax.set_ylabel('Scores (in %)')
ax.set_title('Scores of Classifiers')
ax.set_xticks(index + bar_width / 2)
ax.set_xticklabels(('F1', 'Accuracy', 'Precision', 'Recall', 'ROC AUC'))
ax.legend(bbox_to_anchor=(1, 1.02), loc=5, borderaxespad=0)
fig.tight_layout()
plt.show()

37
References

 S. ChandraKala1 and C. Sindhu2, “OPINION MINING AND SENTIMENT

CLASSIFICATION: A SURVEY,”.Vol .3(1),Oct 2012,420-427
 G.Angulakshmi , Dr.R.ManickaChezian ,”An Analysis on Opinion Mining: Techniques and
Tools”. Vol 3(7), 2014 www.iarcce.com.
 Callen Rain,”Sentiment Analysis in Amazon Reviews Using Probabilistic Machine
Learning” Swarthmore College, Department of Computer Science.
 Padmani P .Tribhuvan,S.G. Bhirud,Amrapali P. Tribhuvan,” A Peer Review of Feature
Based Opinion Mining and Summarization”(IJCSIT) International Journal of Computer
Science and Information Technologies, Vol. 5 (1), 2014, 247-250 ,www.ijcsit.com.
 Carenini, G., Ng, R. and Zwart, E. Extracting Knowledge from Evaluative Text.
Proceedings of the Third International Conference on Knowledge Capture (K-CAP’05),
2005.
 Dave, D., Lawrence, A., and Pennock, D. Mining the Peanut Gallery: Opinion Extraction
and Semantic Classification of Product Reviews. Proceedings of International World Wide
Web Conference (WWW’03), 2003.
 Zhu, Jingbo, et al. "Aspect-based opinion polling from customer reviews." IEEE
Transactions on Affective Computing, Volume 2.1,pp.37-49, 2011.
 Na, Jin-Cheon, Haiyang Sui, Christopher Khoo, Syin Chan, and Yunyun Zhou.
"Effectiveness of simple linguistic processing in automatic sentiment classification of
product reviews." Advances in Knowledge Organization Volume9, pp. 49-54, 2004.
 Nasukawa, Tetsuya, and Jeonghee Yi. "Sentiment analysis: Capturing favorability using
natural language processing." In Proceedings of the 2nd international conference on
Knowledge capture, ACM, pp. 70-77, 2003.
 Li, Shoushan, Zhongqing Wang, Sophia Yat Mei Lee, and Chu-Ren Huang. "Sentiment
Classification with Polarity Shifting Detection." In Asian Language Processing (IALP),
2013 International Conference on, pp. 129-132. IEEE, 2013.

38