Sentimental Analysis using Dictionary and Machine

Learning Approach
Manish Dwivedi Ishika Chaudhary line 1 : Sansar Singh Chauhan
Department of Computer Science and Department of Computer Science Department of Computer Science and
Engineering and Engineering Engineering
GL Bajaj Institute of Technology and GL Bajaj Institute of Technology GL Bajaj Institute of Technology and
Management(GLBITM) and Management(GLBITM) Management(GLBITM)
Greater Noida Greater Noida Greater Noida
[email protected] [email protected]  [email protected]

line 2: Satya Prakash Yadav

Department of Computer Science
and Engineering
GL Bajaj Institute of Technology
and Management(GLBITM)
Greater Noida
 [email protected]

Abstract— Nowadays we can see a significant combination of machine learning and natural
surge in user-generated material on the web as a result language processing (NLP). Sentiment
of enhanced digitization, which gives people's analysis is a technique for analysing the
thoughts on many themes. The computer study of emotions represented in a text. The
assessing people's sentiments and views regarding an computer analysis of human’s views,
entity is known as sentiment analysis. What do people feelings, emotions, and attitudes about
think? How do you feel about a specific topic? things like products, services, issues, events,
Bringing together computer science researchers, themes, and their qualities is known as
Computing linguistics, data mining, psychology, and sentiment analysis or opinion mining (Liu
even sociology are just a few examples. Sentiment 2015). In view of result, sentiment analysis
analysis is a text mining approach that automatically may be used to track public attitude about a
analyses text for the writer's sentiment using machine certain entity and generate useful
learning and natural language processing (NLP) information. This form of information may
(positive, negative, neutral, and beyond). Positive, also be utilized to comprehend, explain, and
Negative, and Neutral comments may be readily
forecast social processes (Pozzi et al. 2017).
identified using powerful machine learning
Sentiment analysis is critical in the business
algorithms. There are various types of sentimental
sphere since it allows companies to develop
analysis. The main feature of emotional is that it
classifies the polarity in text data. The number of strategy and acquire insight into client
smart phones is expanding in lockstep with the growth opinion on their goods. Understanding the
of the internet. The contemporary Internet allows consumer is becoming increasingly vital in
millions of individuals all over the world to connect today's customer-oriented company culture.
with one another and share their ideas and opinions
via email, social networking websites like Twitter,
Facebook and other means. It is the cheapest and most
convenient method of interacting with others. There is
a lot of text material on this social networking site.
These text data may be utilized to analyse public
sentiment on certain issues, as well as the emotion
exhibited on any online platform. In this paper we are
going to review and compare the Traditional
Dictionary Based Approach and Machine Learning
with text classifier which is trained with the dataset of
U.S. Airline under first propose and second propose.
Keywords: Data Pre-Processing, sentiment analysis, NLTK, Figure.1. Flow chart of Sentimental Analysis
matplotlib, Long-short-Term-Memory (LSTM), binary text classifier,
Pandas, TensorFlow.
Fig.1 above demonstrate the flow chart of
Introduction
the steps involve in building the sentimental
One of the industry's most popular analysis model using dictionary based
initiatives, every customer-facing industry approach ,According to the figure first tweet
(retail, telecommunications, banking etc.) is is taken using scrapper as input text for data
interested in determining whether its processing and sentence is broken into
consumers have favourable or negative smallest unit which is word or tokens
feelings about them. Python sentiment followed by removal of stop words to
analysis is a way for examining a piece of transform it into list which contains words
text and determining the hidden sentiment. which are useful in performing sentimental
This is accomplished through the use of a analysis and finally emotion is classified
using the Dictionary which contains words
as key and emotion as value of the key.
Fig.2 is the Pictorial Representation of
The Internet has revolutionized the way the types of basic emotion present in the text
individuals share their thoughts and ideas or words in a conversation which are
nowadays. It is currently mostly Positive, Neutral and Negative.
accomplished through blog entries, internet
forums, product review websites, social
media, and other similar mediums. Millions The three levels of sentiment categorization are
of people use social networking sites like Document level, Sentence level, and Aspect or
Facebook, Twitter, Google Plus, and others feature level. The objective at the document level is
to express their emotions, discuss ideas, and to categorize the entire document into a good or
share viewpoints about their daily lives. We bad category. Sentence level sentiment
receive an interactive media through online categorization divides sentences into three classes:
communities, where consumers utilize positive, negative, and neutral. The polarity of each
internet forums to educate and influence word in a sentence is decided first, followed by the
others. In the form of tweets, status updates, overall mood of the statement. The sentiment
blog posts, comments, and reviews, social classification at the aspect or feature level finds and
media generates a vast amount of sentiment- extracts product attributes from the source data
rich data. Furthermore, social platform before categorization. Machine learning-based
allows businesses to engage with their sentiment analysis and dictionary-based sentiment
consumers for the purpose of advertising. analysis are the two most used techniques to
People make a lot of decisions based on sentiment analysis. To categorize text, a machine
user-generated material found on the learning-based approach uses a classification
internet. For example, before making a algorithm such as a support vector machine or a
decision to buy a product or use a service, neural network. To identify polarity, a dictionary-
people will study it online and discuss it on based technique employs a sentiment dictionary
social media platforms. End-user-generated comprising opinion terms and matches them to the
content is simply too huge for a normal user data. hey give opinion words sentiment scores that
to study. As a result, there is a requirement describe the Positive, Negative, and Objective
to automate. Textual information retrieval scores of the words in the dictionary.
strategies are primarily concerned with
processing, finding, and interpreting the
factual information available. Although facts
are objective, there are certain textual I. LITERATURE REVIEW
components that represent subjective traits.
Opinions, feelings, assessments, attitudes,
and emotions are the most common contents There has already been a lot of study
in Sentiment Analysis (SA). Because of the done on sentiment analysis in the past. The
tremendous proliferation of existing content most recent study in this area focuses on
on the internet via blogs and social media doing emotional analysis on any type of
platforms, it presents numerous difficult text, sentence, paragraph, or someone's
chances for developing new applications. voice, with the majority of the data coming
For example, using SA, it is possible to from social media platforms such as
forecast recommendations of goods Facebook, Twitter, and Amazon. Emotional
provided by a recommendation platform by analysis research, in particular, is focused on
considering criteria like positive or negative machine learning algorithms, with the goal
comments about such products. Sentiment of determining whether a given text
analysis includes a variety of tasks such as encourages or opposes recognizing text
sentiment extraction, sentiment divisions. In this part, you'll get an in-depth
classification, subjectivity categorization, look at one of the most useful research
opinion summarization, and opinion spam activities: sentiment analysis. The following
detection, to mention a few. Its goal is to are some examples of research in sentiment
examine people's feelings, attitudes, views, analysis using various techniques:
and emotions regarding things including P. Pang, L. Lee, S. Vaidyanathan, and others:
products, people, subjects, organizations, "Thumbs up?" by P. Pang, L. Lee, and S.
and services. Vaidyanathan. Proc.ACL-02 conference on Empirical
approaches in Natural Language Processing, vol.10, pp.
79-86, 200[10].
They were the pioneers in the area of
sentiment analysis. Their major goal was to
categorize material based on overall
sentiment rather than simply topic, for
example, good or negative movie reviews.
They use a movie review database to test
machine learning algorithms, and the
findings show that these algorithms
outperform human-made techniques. They
employ Nave-Bayes, maximum entropy, and
Support vector machines as Machine
Learning Algorithms. They also end by
Figure 2. Segments of Sentimental Analysis looking at a variety of characteristics that
make sentiment categorization difficult. Machine learning algorithms employ
They reveal the root of sentiment analysis is computer approaches to "learn" information
supervised machine learning algorithms. directly from data rather than depending on
a model.
“NLTK: the Natural Language Toolkit," in
Proceedings of the ACL-02 Workshop on
Effective Tools and Methodologies for
Teaching Natural Language Processing and
Computational Linguistics, vol. 1, pp. 63-70, Machine Learning Approach is classified mainly
2002 E. Loper and S. Bird [11] into four parts for Sentimental Analysis:
The Natural Language Toolkit (NLTK)
is a collection of software modules,
structured files, tutorials, problem sets, Supervised Learning: Supervised
statistical functions, machine learning learning is one of the methods of
classifiers that are ready to use, sentimental analysis that involves training a
computational linguistics courseware, and computer system on input data that has been
labelled for a certain output.
other resources. One of NLTK's core tasks is
natural language processing, which entails Decision Tree Classifier: In the decision
assessing human language data. Corpora are tree classifier, each node in the tree
provided by NLTK and are used to train represents a test on an attribute, and each
classifiers. Developers replace old branch descending from that node represents
components with new ones, programs get one of the property's potential values.
more organized, and datasets produce more
complex outputs. Linear Classification: A linear classifier
is a model that uses a linear combination of
The Natural Language Toolkit (NLTK) explanatory factors to categorize a set of
is a collection of software modules, data points into a discrete class.
structured files, tutorials, problem sets,
Support Vector Machine (SVM):
statistical functions, machine learning
Support-vector machines are supervised
classifiers that are ready to use, learning models with related learning
computational linguistics courseware, and algorithms for classification and regression
other resources. One of NLTK's core tasks is analysis in machine learning.
natural language processing, which entails
assessing human language data. Corpora are
provided by NLTK and are used to train
classifiers. Developers replace old
components with new ones, programs get
more organized, and datasets produce more
complex outputs.
O. Almatrafi, S. Parack, B. Chavan, and others [12]
According to the researcher Sentiment,
Sentimental Analysis, is the process of
extracting a sentiment from a text unit from
a specific location using Natural Language
Processing (NLP) and machine learning
approaches. They look at a variety of Figure 3. Flow Chart of Machine Learning
Approach
location-based sentiment analysis
applications using a data source that permits
data to be obtained from numerous
Above figure.3 demonstrate the different ways
locations. easily. A script may easily access of using Machine Learning Approach for
a feature of Twitter called tweet location, sentimental Analysis
which allows data (tweets) from a given
location to be obtained for the aim of
detecting trends and patterns. The following
2. Lexicon-based Approach
illustration aims at providing an insight into
more popular algorithms used in sentiment The Lexicon-based technique assesses a
analysis: document by aggregating the sentiment
ratings of all the terms in the content using a
pre-prepared sentiment lexicon. A term and
1. Machine Learning Approach its related sentiment score should be
included in the sentiment lexicon.
Machine learning is a data analytics
technology that trains computers to learn
from experience in the same way that people
and animals do. Machine learning
algorithms employ computer approaches to
"learn" information directly from data rather
than depending on a model. Machine
learning is a data analytics technology that
trains computers to learn from experience in
the same way that people and animals do.
Fig.4 flow chart of Lexicon Based Approach

Above fig.4 demonstrate the various case used in

the lexicon Based Approach for Sentimental Analysis

Corpus Based Approach: The corpus-

based approach to language instruction is
based on genuine and authentic occurrences
of language as it is spoken, written, and used
by native speakers in a variety of settings.
Dictionary Based Approach: Dictionary- II. METHODOLOGY
based sentiment analysis is a computational
approach to measuring the feeling that a text
conveys to the reader. In the simplest case, 1) Dictionary Based Approach:
sentiment has a binary classification:
positive or negative, but it can be extended We reviewed the strategy used in
to multiple dimensions such as fear, sadness, performing the execution of the study and
anger, joy, etc. the algorithm utilized in the Sentimental
Analysis with using scrapper to extract
tweet for text data. System's process flow
Table 1: Contrast Study of Various Sentimental chart. is given below:
Analysis Techniques.

Figure 5. Methodology for Sentiment

Analysis using NLP
There are various steps involved in the
Methodology of the first propose from the Text
data to sentimental analysis of that text data. As
we discussed above that Text data is taken as
input from snscrape which is scrapping tool and
Data is processed after breaking big sentence into
individual words or tokens and removing the
stop-words or unnecessary word from that list,
the next step is Polarization which involves the
nltk or natural language processing to give the
score of the sentiment under class like pos, neg,
neu, compound using SentimentIntensityAnalyzer
function and the final step is to analyze the
sentiment based on the score.

Text data from snscrape [SNS]

Text data is taken from Scraper for
social networking services, snscrape
(SNS). It scrapes user profiles,
hashtags, and tweets and returns the
items found, such as related postings.

Data Pre-Processing
 The unanalyzed data is handled in Figure 6. Methodology Of
preprocessing for feature extraction. It Sentimental Analysis using Binary text
is further broken down into the classifier.
following steps: The steps involve under the
a) Tokenization: A phrase is broken propose-4 in the fig.4 are as
down into words by removing follows, firstly csv file or dataset is
white spaces, symbols, and spe- taken from the Kaggle which have
cial l characters.
more than 14,000 tweets of us-
airline sentiment and then huge
dataset is converted and cleaned
b) Stop words removal: some using pandas which make dataset of
words like article, adjective etc. more than 6 column into 2 column
are removed using NLTK corpus which are text and airline sentiment
library which does not have any and binary text classifier algorithm
kind of emotion. is used with L.S.T.M to predict the
class.

c) Case Normalization: The entire documents

are converted into lower-case. Dataset: Contains:

More than 14000 tweet data

samples are included in the
Data Polarization: The collection, which are classified into
orientation of the stated emotion is three groups: Positive, Negative, or
determined by the element's Neutral.
sentiment polarity, which
determines whether the text
communicates the user's positive,
negative, or neutral feeling toward
the entity in question. The main
motive of sentimental analysis is to
examine a body of text in order to
determine the viewpoint
communicated. We usually measure
this feeling with a positive or
negative polarity value. The polarity
score's sign is usually used to assess
whether an emotion is positive,
neutral, or negative.

2) Machine Learning Based Approach: Figure. 7 flow of Data preprocessing and cleaning

The methodologies utilized in Figure.7 utilized the US Airline

the execution of the study and the Twitter Dataset, which comprises
Binary text Classifier algorithm over 15,000 tweets, in this work.
that are employed in the class The dataset initially contains
prediction using sentiment analysis parameters such as twitter id, airline
are explained in this model. The sentiment, text, airline sentiment,
flow chart of the approach used in confidence, airline, name, which are
this article is shown below. then reduced to two attributes,
airline sentiments and text, using
feature extraction. In this paper, we
used a binary text classifier that
gives 0 and 1 for negative and
positive classes for column airline
sentiment.

Dataset after feature Extraction

 supplied to them. This association
is stored in the form of a dictionary
in the tokenizer. word index
attribute. Using the text to sequence
Data Pre-Processing and Cleaning:
() function, replace the words with
their allocated numbers.
we have used binary text
classifier which only takes two The sentences in dataset are not all the
class as we don't need neutral same length. To make the sentences equal in
reviews from the dataset therefore length, use padding.
it has been removed from the
dataset.
Binary Classifier using LSTM:

In our machine learning model for sentiment analysis, we

employ LSTM layers. Our model has three layers: an
embedding layer, an LSTM layer, and a Dense layer in the
end. We used the Dropout mechanism in-between the LSTM
layers to minimize overfitting.

This dataset's labels are

categorical. Only numeric data is
understood by machines. So, using
the factorize () function, transform
the category data to numeric values.
This gives you an array of numeric
numbers as well as a category Figure.8 Long- short-term neural
index. network

Figure.8 explains about the Mathematical

working of the Long-short-Term Neural
Network
The 0 signifies good feeling and
the 1 represents negative sentiment,
as you can see. Now comes the
most important part of python
Long Short-Term Memory
sentiment analysis. The input text
Networks are abbreviated as LSTM.
should be transformed into
It's a Recurrent Neural Networks
something that our machine
learning model can understand. variation. Recurrent Neural
Therefore, the text has been Networks are typically used to
converted into vector embeddings process sequential input like text
array. Word embeddings are a and audio. The meaning of each
lovely method of displaying the word and associated computations
relationship between words in a (known as hidden states) are usually
text. To do so, we assign a unique kept while constructing an
number to each of the unique embedding matrix. If a word's
words, then replace the word with reference is utilized after 100 words
the assigned number. in a text, then RNNs cannot keep all
of these computations in their
memory. RNNs are unable to learn
these long-term dependencies for
Now, before proceeding ahead this reason. One of the
in python sentiment analysis project regularization techniques is
let’s tokenize all the words in the dropout. It is employed to prevent
text with the help of Tokenizer. In overfitting. We drop some neurons
tokenization, we break down all the at random in the dropout
words/sentences of a text into small mechanism. The layer accepts a
parts called tokens. The fit-on texts value between 0 and 1 as an
() function establishes a link argument, which reflects the
between the words and the numbers likelihood of dropping the neurons.
 This results in a stable model that III. RESULT AND DISCUSION
avoids overfitting.
Below table 2 illustrates the comparison and
similarities between two proposed approaches
in this paper using strength and challenges
given below:

Class Prediction:

With a batch size of 32 and a

validation split of 20%, train the
sentiment analysis model for 5
epochs on the whole dataset.

We successfully constructed a sentiment

analysis model in Python. We created a binary
text classifier in this machine learning project
that divides tweet sentiment into positive and
negative categories. On validation, we got
more than 94 percent accuracy. Let's use
matplotlib to plot these metrics. Matplotlib is a
visualization tool that uses a low-level graph
plotting toolkit written in Python.

We have defined a function to predict sentiment that

takes the input as a sentence and classifies it into one of
the two-class that is Positive and Negative. The accuracy
of the model is approximately 94.0 percent. as we see that
the test_sentence1 which seems to be negative is also
predicted negatively by our text classifier model and .
text_sentence2 which seems to be positive is also
predicted correctly by our text classifier model.
Figure .9 Accuracy Function of the Model
In Figure.9 the orange line in the graph shows the training Whereas, according to the
accuracy is approx. to 96 percent and testing accuracy fig.11 When the identical tweet is
approx. to 94 percent respectively. fed into the text classifier model, it
produces a negative classification.

IV. RESULT AND DISCUSION

This work is divided into two
sections under the first propose and
second propose. The first part of the
project focuses on sentimental
analysis using a dictionary-based
technique, while the second portion
focuses on sentimental analysis
using a machine-learning approach.
After comparing both, it was
determined that the second propose
advised the correct sentiment for a
particular tweet with a testing
accuracy of 94%. our second
proposal, a machine learning
technique, has certain limitations.
Therefore, if the provided tweet
contains terms that are not included
in the dataset, the class prediction
may be incorrect for few tweets as
Figure.10 loss function of the Model input.

Now, we have compared the

both Dictionary based approach
and text-classifier approach for the
given tweet and the following
results are obtained:

Figure 10: output of first propose

V. References
Figure 10 shows the result of a
sentiment analysis for the
following tweet using Dictionary 1. G. Vinodhini and RM. Chandrasekaran,
based Approach as a positive class. “Sentiment Analysis and Opinion Min-
ing: A Survey”, Volume 2, Issue 6, June
2012 ISSN: 2277 128X International
Journal of Advanced Research in Com-
puter Science and Software Engineering

2. A. Pak and P. Paroubek. „Twitter as a

Corpus for Sentiment Analysis and Opin-
ion Mining". In Proceedings of the Sev-
enth Conference on International Lan-
guage Resources and Evaluation, 2010,
pp.1320-1326

3. Madhoushi, Z. Hamdan, A. R. and Zain-

Figure 11:output of second propose uddin, Sentiment Analysis Techniques
 in Recent Works,” Science and Informa- Language Toolkit”, Proc. ACL-02 Work-
tion Conference (SAI), 2015. IEEE shop on Effective tools and method-
ologies for teaching natural language
processing and computational linguis-
4. Singh and Vivek Kumar, ―A clus- tics,vol. 1,pp. 63-70, 2002
tering and opinion mining ap-
proach to socio-political analysis
of the blogosphere”, Computa-
tional Intelligence and Comput-
ing Research (ICCIC), 2010 IEEE
International Conference

5. Madhoushi, Z. Hamdan, A. R. and Zain-

uddin, Sentiment Analysis Techniques
in Recent Works,” Science and Informa-
tion Conference (SAI), 2015. IEEE

6. Jha, I. Manjunath, N., Shenoy, Venu-

gopal, K. R. and Patnaik, L., “HOMS:
Hindi Opinion Mining System,” IEEE 2nd
International Conference on Recent
Trends in Information Systems (RTIS) M.
2015.

7. E. Loper and S. Bird, “NLTK: the Natural