Sentiment Analysis of a Product based on User

Reviews using Random Forests Algorithm

Dr. Shailendra Narayan Singh

Twinkle Sarraf
Computer Science and Engineering Computer Science and Engineering
Amity School of Engineering and Technology Amity School of Engineering and Technology
Amity University,Noida, India Amity University,Noida, India
[email protected] [email protected]

Abstract— After many sentiment analysis as well as many types customers consider, what sort of things individuals truly like
of methods classify the reviews that is based on test data and and don't like, item, benefit which may truly help
reviewer’s ratings which uses training. , after reading reviews it organisations to settle on choices. These days the greater part
is seen that star rating of reviewer do not always give a precise of the general population don't purchase things without
measure of his sentiment. This paper primarily focuses on
making some examination of the item over the web,
analyzing customer reviews from the e-commerce space. Upon
surveying popular e-commerce websites it can be observed that individuals check for the item audits/reviews and after that
in several instances the product rating given by a customer is not settle on their choices. Back in the time when organisations
consistent with the product review written by him/her. The required the general population or buyers' conclusions when
problem is made complex by the fact that there is no standard they need the conduction of opinion surveys that maybe costly
scale to measure the rating that the user gives and the rating of and additionally tedious and will require human resource.
the product are instinctive to the customers’ view. In several This presents challenges which may not be easily addressed
cases it is seen that a product is rated 4 out of 5. However, the by the aid of simple text classification tactics. Thereupon,
reviews detail that the customer’s experience with the product is there is a need to incorporate techniques for classifying
not favourable. Indeed, text reviews are a true picture of the
opinions into a simplified text classification tool, or to develop
product. To get rid of this problem, the stated system will give a
boolean result i.e. whether the product is good or bad and the systems that will be able to accurately analyze and classify
user does not need to read all the reviews to analyze the product. sentiments in text. A kind of contextual mining which helps to
identify and extract subjective information or subjective data
Keywords used — Sentiment analysis, product reviews, is called sentiment analysis. This type of extraction helps a
random forest classifier, bag-of-words business to understand their brand’s social sentiment or
product or services when they monitor online conversation.
I. INTRODUCTION This need for analyzing the sentiments has risen in recent
years due to the application of sentiment analysis in varied
Formally this sentiment analysis has been referred to
areas such as business intelligence, research, public relations,
as a kind of analysis using natural language processing, some
e- governance and web search Other factors that tends to
sort of computational linguistics as well as text mining. When
develop in the increased rate in sentiment analysis are
a n individual makes a decision, it may be the decision or
discussed below -
thought influenced by others influence. Moreover, internet
• The rising of methods of machine learning and
provide a forum for this. We can take the example of
retrieval of particular data.
flipkart.com customer feedback system for the rating of
products that they receive from Flipkart and at the same time • The availability of data sets in the field of machine
allows other customers to make a more informed decision by learning algorithms which must be trained, in general
making the ratings available to other customers to review there must be culmination of websites related to
before they make a purchase decision. Almost every business review-aggregation.
organisation today are in rush to realize that whether • The realization of difficulties that are offered by the
individuals like their items and administrations, what do intelligence and commercial applications.

c
978-1-7281-2791-0/20/$31.00 2020 IEEE 112

Authorized licensed use limited to: Somaiya University. Downloaded on November 06,2023 at 17:17:30 UTC from IEEE Xplore. Restrictions apply.
 Figure1: Google Trends data showing the relative popularity
of search strings “Sentiment analysis” and “Costumer
feedback”
Figure2: A typical sentiment analysis model
II. STRUCTURAL DESIGN OF OPINION MINING
III. RELATED WORK
A typical type of sentiment analysis model is given in
the figure 1. This model processes the reviews that it takes as Sentiment Analysis under the topic of micro-
input by three important steps: Preparation of data, review blogging is a topic which is recent in the research topics and
Analysis as well as Classification of Sentiment. This model there can be even more research related to this topic. A large
produces output for classification of reviews. amount of related work related to reviews, documents is done
on sentiment analysis of above factors including general phase
A. Data Preparation level sentiment analysis. We may take naive bayes and
The pre-processing of data and cleaning on the reviews is supporting vector machines being supervised learning
performed by the data preparation step for the continuous machines as the best results but for the supervised approach,
analysis. Some of the frequently used steps of pre-processing but the manual labeling required is very expensive.
contains of contents such as '.',',' etc. and HTML tags as well Approaches that are applied are semi-supervised and
as removal of irrelevant data from the reviews for sentiment unsupervised and there could be many more improvements.
analysis, such reviewers’ name and review date[3]. Many different researchers who compare their result to the
base-line other performances. These proper and formal result
B. Review Analysis comes after these comparison for the selection of best and
most efficient features and classification technique. Therefore,
For extracting all the relevant and interesting information like the proper performances comes after the comparison.
the opinions and analyzing the linguistic characteristics of Among many ways, of the sediment analysis, in this
reviews, the second step of review analysis is performed. This paper we are going to make a focus with the use of machine
step before extracting suggestions and product characters, learning approaches on sentiment analysis.
processes the opinion by applying different types of tasks
which is computational linguistic in nature. The opinion IV. PROPOSED APPROACH
analysis procedure then steps to extract opinion from
processed reviews. In this work the live data of certain e-commerce sites
will be collected using their respective URLs. The data
C. Sentiment Classification collected will include customer reviews about the product.
The system will crawl the URLs in order to find the opinions.
The basic two types of approaches to classifying reviews are: Here we are scrapping the data from internet so that we can
1) Machine learning approach have exact opinions as per user requirement.The data crawled
2) SO approach from the website will be parsed in order to extract the reviews,
We will discuss about machine learning approach for which will be subjected to analysis and processing.
sentiment classification in this paper. This type of This system is different from the existing system in the sense
approach is somewhat similar to classifying top into that no pre-existing data sets are used but the live data
positive sentiment class and negative sentiment class. currently running on the sites i.e. the latest reviews and ratings
The reviews are then stepwise broken down in phases of the users are analyzed to calculate the final boolean result
or in the words, represents the review in the form of a of the product. This system analyses the text based opinion of
document vector defining the opinions which is the user using natural language processing techniques and puts
based on document vectors. to use the word model’s bag. A popular feature which depicts
simplicity as well as good performance is the bag-of-word.
The model helps to represent the text with no connection of

10th International Conference on Cloud Computing, Data Science & Engineering (Confluence) 113

Authorized licensed use limited to: Somaiya University. Downloaded on November 06,2023 at 17:17:30 UTC from IEEE Xplore. Restrictions apply.
the words with each other. This model is one of the popular BeautifulSoup pulls the data of HTML and XML files out.
models and is very useful in the process of sediment analysis Using your favourite parser to give out the various ways of
and many other researches. The best and simple method to search, navigate and modify the parse tree.
include this model in our classifier is by using uni-grams. A
collection of particular word in the text to be classified , where When the user enters the product's name, the product is
the word once used is not affected by usage of other word . goggled and web crawling is done. The URL of the search
The Bag of Words model tells about the vocabulary used in result page must be known before any code performed. On
the sentences as many times but counted only once as: any search on google, the browser’s address bar contains the
URL:
Sentence 1: "The bell fell " https://www.google.com/search?q=SEARCH_THE_TERM_H
Sentence 2: "The audible bell once fell in the well" ERE the page is then downloaded after the search. Finally, the
This makes our vocabulary as : web browser module is used to open the browser tabs for that
{ the, bell, fell, in, audible, the, once, well, ate, and } link. Now Flipkart has been appended to the url so as to search
The bag of words can be analyzed by the multiple usage of a for the product on flipkart. After this again beautiful soup is
word occurred in each sentence. For example "the", "bell" used to crawl the links on the web page. The first link is
"fell"" each appear once. crawled finding the data-href under the r class. After having
To limit the limitation of feature vector, a particular size of crawled the link, next task is to scrap the reviews. The reviews
vocabulary must be chosen. The most frequent 5000 words are have been scraped by again using beautiful soup. The
used here. contents/ reviews are scrapped and stored in comma separated
The sentiment analyzer of reviews after cleaning of reviews files along with the review id. The tab separated input file is
and implementing Bag of Words model for uses Random fed to the code which implements Random Forests algorithm.
Forest Classifier for classification. This section focuses on Forests are constructed for training dataset. The constructed
various forests classifier, as well as the impact they make forests will be traversed to arrive at the sentiment of each
related to accuracy and other features. The first paper made review in input file. If the number of positive sentiments for
that focuses on the ensemble of decision trees that was an input file is >= 50% of the total number of input reviews,
composed of multiple tree combination were random forest. then the product is recommended. Otherwise, it is not.
Problems like noise or outliners may occour that may affect
the result in the single tree classifier, whereas random forest is The output is also stored in the form of comma separated file
much robust to noise as well as the provided randomness. Two containing id, review and sentiment as columns.
types of randomness namely bagging and bootstrapping are
the main concept of random forest classifier. Initially the raw data is loaded. The sentiment analysis is done
using Natural language processing and after that vader which
is a part of NLTK module can be used . It uses a lexicon of
words to find negative and the positive ones. It also analyses
Algorithm of Random Forest the sentiments to determine the sentiment scores. After which
vader returns four values for each text:
Input: No. of trees= B, Training Data = N, Total Features, f =
Subset of Features z A neutrality score
Output: for input data, bagged class level. z A positivity score
z A negativity score
1.Analysing each tree in forest B: z An overall score that summarizes the previous score
a) Selection of a bootstrap sample S of the size N from
training data. Start
b) creation of tree Tb repeating recursively the following
steps:
i. randomly choosing f from F.
User enters the name of the product in the text box and
ii. Selection of best from F.
clicks on analyze button or press enter.
iii. Splitting of the node.

2. After creating B trees, the instance of test will be passed to

each tree as well as there will be assignment of class label o Web crawling and Web Scraping
the basis of majority of votes. Web page of flipkart is crawled and reviews for the
product are scraped.
V. IMPLEMENTATION AND RESULT

Data in the form of raw reviews have been scraped

from Flipkart.com using BeautifulSoup library of python. The reviews are cleaned and saved in a file along with
an id.

114 10th International Conference on Cloud Computing, Data Science & Engineering (Confluence)

Authorized licensed use limited to: Somaiya University. Downloaded on November 06,2023 at 17:17:30 UTC from IEEE Xplore. Restrictions apply.
 Figure5: File Bag_of_words .csv after implementing Bag-of-
words model and random forest algorithm

The bag of words model and random forest are

implemented. The result in the form of 0(negative) and
1(positive) is saved in another .csv file.

The file is then opened and the positive and negative

reviews are counted and the percentage of both is
calculated.

This percentage is then displayed to the user through

user interface in the corresponding label.
Figure6: User Interface displaying the final result for entered
product
End
Figure3: Control Flow Diagram of the system
V. CONCLUSION

Sentiment analysis contains some sentiments whose

classification of text must be dealed with. This paper consists
of typical sentiment analysis model which comprises of most
importantly three core steps that is named as data preparation,
review analysis and sentiment classification as well as It
describes representative techniques. Sentiment analysis
involves the research in the field of text mining and
computational linguistics. It was a good option to attract the
significant research attention since last some years.
The issue of rating – review disparity is addressed by
this tool. It gives boolean review based on reviews and not on
ratings which avoids the problems mentioned earlier. The
reviews are extracted from e-commerce site Flipkart.com so
that the reviews are accurate.. It gives boolean review of
product. Therefore, it will be easier to decide. The count of
negative as well as positive reviews are also displayed to show
Figure4: File Review.csv where the scraped reviews are saved
the precision of recommendation. The boolean result whether
the product is recommended or not is displayed in user
friendly interface.
.

VI. FUTURE ENHANCEMENTS

In the domain of micro-blogging, the working of

sentiment analysis is still developing an it is far from
complete. We should get some ideas to explore in the future
development and performance improvisation. The project can
be enhanced to take input as the Url of the product.

The reviews can then be extracted directly from the

entered url. The tool currently displays the sentiment based on
both product reviews and seller reviews. It can be further
enhanced to display sentiment based on product reviews and

10th International Conference on Cloud Computing, Data Science & Engineering (Confluence) 115

Authorized licensed use limited to: Somaiya University. Downloaded on November 06,2023 at 17:17:30 UTC from IEEE Xplore. Restrictions apply.
seller reviews separately. It can be web hosted with a different [5] Hitesh Parmar, Sanjay Bhandari, Glory Shah (2014, July).
Sentiment mining of Movie Reviews using Random Forest with Tuned
database format. It can be extended for some more e- Hyperparameters. Presented at: International conference on Information
commerce sites. The tool currently analyses reviews only from Science, Kerala. [Online]
Flipkart. It can further be enhanced for more than one e- [6] Bing Liu. (2010). Sentiment Analysis and Subjectivity.
commerce sites like Amazon, eBay etc. so looking for just our Handbook of Natural Language Processing, Second Edition. (editors: N.
topic and focusing on the uni-gram, further exploring bi-grams Indurkhya and F. J. Damerau)
and tri-grams. As when used with bi-grams, uni-grams usually [7] L. Breiman, Random forests, Machine Learning, vol. 45. Issue 1,
pp. 5-32G. Eason, B. Noble, and I. N. Sneddon, “On certain integrals of
enhanced I it’s performance. Lipschitz-Hankel type involving products of Bessel functions,” Phil.
Trans. Roy. Soc. London, vol. A247, pp. 529–551, April 1955.
(references)
REFERENCES [8] Apoorv Agrawal, Boyi Xie, Owen Rambow, Sentiment analysis of
Twitter data. Columbia University, New York, NY 10027 USA.[Online]
[9] Mika V. Mantyla Daniel Graziotin, Miikka kuutila, The evolution
of sentiment analysis, ITEE , University of Stuttgart, Finland.
[1] Andranik Tumasjan, Timm O. Sprenger, Philipp G. Sandner and [10] Dipankar Das, Souvick Ghosh and Tanmoy Chakraborty .
Isabell M. Welpe. Predicting Elections with Twitter: What 140 Determining sentiment in citation text and analyzing its impact on the
Characters Reveal about Political Sentiment. In Proceedings of AAAI proposed ranking index. Jadavpur University, Kolkata. (references)
Conference on Weblogs and Social Media (ICWSM), 2010.
[2] Bo Pang and Lillian Lee. (2008). Opinion mining and Sentiment
analysis. Foundations and Trends in Information Retrieval Vol. 2, Nos.
1–2. Pages: 1–135
[3] Stephen C. F. Chan, Cane W. K. Leung* Sentiment Analysis of
Product Reviews
[4] B. Pang , L. Lee, and S. Vaithyanathan, Thumbs up?:sentiment
classification using machine learning techniques, Proceedings of the
ACL-02 conference on Empirical methods in natural language
processing, vol.10, 2002, pp. 79-86

116 10th International Conference on Cloud Computing, Data Science & Engineering (Confluence)

Authorized licensed use limited to: Somaiya University. Downloaded on November 06,2023 at 17:17:30 UTC from IEEE Xplore. Restrictions apply.