Live Twitter Sentiment Analysis and Interactive Visualizations with PyLDAvis using Streamlit

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 07 | July 2023 www.irjet.net p-ISSN: 2395-0072
© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 1187
Live Twitter Sentiment Analysis and Interactive Visualizations with
PyLDAvis using Streamlit
Sabbineni Lakshmi Gopi Koushik 1, Chinthapatla Navyasri2, Gurram Keerthana3, Chunduru
Jahnavi4
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Sentiment analysis, often known as opinion
mining, is a technique for determining the emotional tone or
attitude indicated in a piece of text, such as a tweet, review,
or news story. As Twitter has many influential users, it
became an important tool for communication and
information exchange on various platforms including
politics, business, and entertainment. This paper provides a
user-friendly online application for 'sentiment analysis of
live Twitter data and interactive visualizations using
pyLDAvis' built on Python's VADER module and the
Streamlit framework. Our application asks users to enter a
topic or phrase of interest and then uses the Twitter API to
stream tweets in real-time. We use VADER to do sentiment
analysis on incoming tweets and illustrate the results in
real-time using various interactive charts and plots. Further,
we implement PyLDAvis to do topic modeling on the tweets
and display the topics and keywords connected with them.
In a dynamic and interactive manner, our application
allows users to explore the emotion and themes of Twitter
conversations connected to their areas of interest.
Key Words: Sentiment Analysis, pyLDAvis, Streamlit
Framework, Vader library.
1.INTRODUCTION
Natural Language Processing (NLP) is a rapidly emerging
computer science topic that has grown in significance over
the past few years. NLP is focused on teaching machines
how to read and process human language in the same way
that humans do. This entails creating algorithms and
computer processes capable of analyzing, interpreting,
and producing human language data such as text or
speech. NLP offers a wide range of practical applications,
including text classification and sentiment analysis, as well
as machine translation and chatbots. Sentiment analysis,
commonly referred to as opinion mining, is an important
method for assessing and analyzing sentiment
represented in text data. Sentiment analysis uses
computer techniques and algorithms to recognize and
classify the emotional tone or attitude indicated in a piece
of text, such as a tweet, review, or news story. Sentiment
analysis is a tool that can analyze client feedback and
evaluations, helping businesses in understanding the
benefits and drawbacks of their products or services. It
can also assist customer care personnel in promptly
identifying and responding to bad feedback or complaints
posted by customers on social media. Amongst the most
difficult things of sentiment analysis is appropriately
recognizing the sentiment expressed in text, particularly
true for ambiguous sentences in which the sentiment may
be unclear. Researchers and practitioners in sentiment
analysis are always attempting to increase the accuracy
and reliability of sentiment analysis approaches by
employing modern machine learning algorithms and
natural language processing techniques. Irrespective of
challenges, sentiment analysis has grown in importance in
the age of big data.
1.1 Problem in Existing System
The problem with the existing system is limited training
data. To learn how to identify sentiment in text, sentiment
analysis systems rely significantly on training data.
However, if the training data is inadequate or polarised,
the system may be unable to identify sentiment in new
material effectively. Another problem of not having a
display of sentiment analysis results is that it can be
difficult for users to quickly and readily understand the
sentiment of the text. Without a visual representation,
consumers may have to go through huge amounts of text
or data to understand the overall sentiment, which can be
time-consuming and ineffective. Patterns and trends in
sentiment analysis data may not always be seen from the
text alone, but visualization can help. A visualization, for
example, may illustrate the frequency of good, negative,
and neutral sentiments across time, allowing viewers to
observe changes in sentiment over time.
2. PROPOSED WORK
The proposed work provides an architecture for live
Twitter sentiment analysis, which is based on a lexicon-
based approach and includes classification issues in real-
time. This model is entirely built on querying real-time
tweets from the Snscrape and pre-processing them into a
corpus of words using a Lexicon-based method, where the
terms are specified.

Fig -1: Proposed Work
3. PROPOSED METHOD
3.1 Data Collection
We must extract the data in order to comprehend
emotions and carry out sentiment analysis. Twitter is used
to collect the data as a consequence. The information from
tweets, comments, and hashtags on Twitter is extracted.
Tags used to extract the data,
 Our task in this study is to deal with data tagging.
Users, tags, and resources are the three entities
that typically comprise a social tagging system.
 Label: #, date, User, Tweet
 Tags we used to extract the data: #Digitalindia,
#DigitalPayments.
Libraries used to extract the data,
 Using the Twitter API or Tweepy, Twitter enables
us to collect the data of any user. The user's
tweets will be extracted as the data.
 Snscrape (SNS) is the acronym for social
networking service scraper. It retrieves the stuff it
has found by scraping information from searches,
hashtags, and user profiles.
 The number of tweets collected is lower while
using tweepy to extract the data. So, we extracted
hashtags, comments, and related posts using
snscrape.
3.2 Data Cleaning
The extracted data is inconsistent. Noisy data requires
more storage and makes sentiment analysis less accurate.
Stop words, hashtags, mentions, and non-English words
can all be found in extracted tweets. Therefore, this data
must be cleaned before performing sentiment analysis and
topic modeling.
3.2.1 Removing special symbols and emojis.
 We use Regex to remove characters like #, and @
as collected tweets also contain mentions.
 Special characters present in the data make them
unsuitable for sentiment analysis.
3.2.2 Removing stop words.
 Stop words are some English words that are not
helpful in sentiment analysis, such as "I," "am,"
"and" and "an." Instead, they result in more
storage and inaccurate analysis.
 Stop words can be eliminated to provide more
attention to meaningful words.
 We use nltk (Natural Language Tool Kit) which is
a suite that contains libraries to process the
textual data.
3.2.3 Remove punctuations.
 In English, punctuation marks are symbols that
provide the sentence additional meaning and are
helpful for grammar. For sentiment analysis,
however, these symbols are useless.
 '!"#$%&'()*+,-./:;<=>?@[]^_`{|}~' • The above
punctuations are removed using string.
punctuation library in nltk.
 In English, the words ‘Digital’ and ‘digital’ are the
same. But, in NLP models, these are treated
differently.
 To avoid this ambiguity, we convert the entire file
into the same case either lower or uppercase.
 Data[]=Data[].apply(lambda x:x.lower())
3.2.4 Tokenization
 In Python, we use the .split() function to divide
the contents of an array or list. In NLP, we
tokenize words similarly by taking into account
white spaces.
 We can use either the .split() method or the
wordpunkttokenizer library of nltk.
3.2.5 Lemmatization
 Lemmatization is the process that reduces a word
to its simple forms. It is used to eliminate
additional tweet noise.

 For Example, the words like loving, and caring
after lemmatization becomes love and care
respectively.
 By lemmatization, we can get the most possible
accurate frequency distributions.
4. IMPLEMENTATION
4.1 Sentiment Analysis
Now, the data is cleansed from all stop words, and
punctuations, and it is in its base form, the data is perfect
for sentiment analysis. It gives accurate results as the
noise is removed from the data. We used VADER as a
modeling approach for sentiment analysis which is a rule-
based sentiment analysis tool designed for social media
posts. It determines the sentiment of a piece of text by
using a lexicon of words and emoticons that have been
graded for their valence (positive, negative, or neutral)
and intensity. VADER can also detect contextual cues such
as capitalization and punctuation, which might influence
the sentiment of a text. VADER's ability to handle rejection
and sarcasm is one of its key advantages. Even though the
word "happy" is positive, VADER would correctly detect
the sentence "I'm not happy with my new dress" as
negative. The VADER lexicon is a collection of words and
phrases that have been graded on two dimensions:
valence and intensity.
 Valence scores range from -1 (extremely
negative) to +1 (extremely positive), with 0
signifying neutral sentiment.
 The intensity score runs from 0 (least intense) to
2. (Most intense).
4.2 Topic Modeling
The process of removing necessary properties from a Bag
of words is called topic modeling. This is significant
because NLP treats each word in the corpus as a feature.
4.2.1 Latent Semantic Analysis (LSA)
By developing a list of concepts that are statistically
related to the documents and terms, distributional
semantics, a subset of latent semantic analysis (LSA), is
used in natural language processing to evaluate the
relationships between a set of documents and the terms
they contain. LSA is mostly used for concept searching and
automatically classifying documents. In order to represent
the text data in terms of these topics or latent qualities, the
latent semantic analysis must be performed.
4.2.2 Latent Dirichlet Allocation (LDA)
Latent Dirichlet Allocation (LDA) is a topic model type
used to classify a document’s text into different categories
based on a certain topic. It constructs a topic-per-
document model and a words-per-topic model using
Dirichlet distributions as the modeling basis. Following
pre-processing operations, the model was trained using
(LSA, and LDA), and the results were then evaluated using
the coherence value. These outcomes demonstrate that the
LDA technique performed better than the LSA technique.
4.2.3 PyLDAvis
A useful tool for visualizing topics from a topic model is
the PyLDAvis module. Use the pyLDAvis cell in the
notebook to execute the interactive dashboard.
Fig -2: Topic Modeling
4.3 Streamlit Deployment
Streamlit is a commonly recognized Python framework for
creating engaging online applications for data science and
machine learning events. Deploying a Streamlit application
entails hosting it on a server and making it accessible
through the internet. Streamlit offers a free hosting
platform called Streamlit Sharing, which allows you to
quickly distribute your Streamlit app. Connecting your
GitHub repository to Streamlit Sharing and delivering your
app with a few clicks makes it simple to deploy your
project.

Fig -3: App overview
The deployed streamlit application receives the input
topic/text and tweet count. If no tweet count is specified,
the default tweet count is 50. It calculates the sentiment
score for a given topic and categorizes it as negative or
positive. It also has manual buttons for producing outputs
like word clouds, pie charts, and interactive dashboards.
5. RESULTS
Fig -4: Firstly, enter the text/topic of the latest news or a
product.
Fig -5: Analysis Performed
As you can see, the total number of tweets is divided into
three categories: positive, negative, and neutral. These are
combined to form the public's concluding view on the
subject. You may explore the extracted text, word cloud,
pie chart, and interactive dashboard by selecting one of
the options listed below.
Fig -6: Extracted data with the fields of cleaned tweets,
positive, negative sentiment, etc. are displayed.
Fig -7: Pie chart of Sentiments
This is illustrated by the pie chart, for Digital India, where
it is noticed that 59.57% of tweets are favorable, 17.02%
are negative, and 23.40% are neutral. According to the
research, the majority of the population is positive about
this initiation.

Fig -8: Word cloud
Presenting the word cloud, which represents the most
repeated terms in a text, with the size of each word
representing its relative frequency or importance.
Fig -9: PyLDAvis Dashboard
Interactive dashboards provide Realtime sentiment trend
visualization and can give insights into how sentiment
evolves over time.
6. CONCLUSIONS
Finally, Live Twitter Sentiment Analysis and Interactive
Visualizations with PyLDAvis Using Streamlit is a powerful
tool for real-time sentiment analysis on social media. The
application analyses the sentiment of tweets using natural
language processing techniques and visualizes the findings
via interactive visualizations.
Users do use PyLDAvis to investigate the subjects
addressed in tweets and acquire insights into the most
popular themes and attitudes. Streamlit offers an easy-to-
use interface for engaging with data, making it simple to
explore and analyze. Overall, Live Twitter Sentiment
Analysis and Interactive Visualizations with PyLDAvis
Using Streamlit might be a useful resource for
organizations and people interested in monitoring and
analyzing social media sentiment. This tool can assist by
offering real-time insights into public sentiment.
REFERENCES
[1] Patil, Shilpa and Lokesha, V., Live Twitter Sentiment
Analysis Using Streamlit Framework (May 25, 2022).
[2] Elbagir, S., & Yang, J. [2019, March]. Twitter sentiment
analysis using natural language toolkit and VADER
sentiment. In Proceedings of the international
multiconference of engineers and computer scientists
(Vol. 122, p. 16).
[3] Jagdale, R. S., Shirsat, V.S., & Deshmukh, S. N. [2016].
Sentiment analysis of events from Twitter using open-
source tool. IJCSMC, 5(4), 475-485.
[4] Prabhakar Kaila, D. [2016]. An Empirical Text mining
analysis of Fort Mcmurray wildfire disaster twitter
communication using topic model. Disaster Advances,
9(7).
[5] Rai, S., S B, G., & Kumar, J. [2020]. Sentiment Analysis
of Twitter Data. International Research Journal on
Advanced Science Hub, 2, 56-61.

Live Twitter Sentiment Analysis and Interactive Visualizations with PyLDAvis using Streamlit

More Related Content

Similar to Live Twitter Sentiment Analysis and Interactive Visualizations with PyLDAvis using Streamlit (20)

More from IRJET Journal (20)

Recently uploaded (20)

Live Twitter Sentiment Analysis and Interactive Visualizations with PyLDAvis using Streamlit