INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
Sentiment Analysis of COVID 19 Tweets using Optimize LSTM  
Model  
Tanzeela Qureshi, Dr. Mohit Singh Tomar and Dr. Ritu Shrivastava  
Department of CSE, SIRT, Bhopal, India  
DOI : https://doi.org/10.51583/IJLTEMAS.2025.1412000110  
Received: 27 December 2025; Accepted: 01 January 2026; Published: 10 January 2026  
ABSTRACT  
The popularity of social media has increased curiosity in psychology, mental health, and human circumstances.  
Twitter and other social media platforms have been utilised for data collection, personality type prediction, and  
sentiment analysis during emergencies. Deep learning techniques examine both positive and negative feelings.  
The retrieval of Covid 19 tweets, both positive and negative, is not perfect, and low accuracy may lead to the  
detection of unidentified tweets from social sites. The objective is to improve the retrieval and comparison of  
pertinent Covid 19 positive and negative tweets in terms of precision, recall, and train/validation accuracy. A  
LSTM is defined by a sequence of layers, with the first layer defining inputs. The ADAM optimization algorithm  
updates weights, and the model is evaluated. The proposed model for sentiment content categorization from  
tweets uses LSTM-ADAM approaches, with a 78% accuracy rate compared to the 76% accuracy of SVM  
methodology. It supports manual participation and accepts text information in any format.  
Keywords: Covid-19, LSTM, ADAM, Accuracy.  
INTRODUCTİON  
Social scientists and psychologists have been interested in the unheard-of proliferation of information on social  
media in order to get a deeper knowledge of psychology, mental health, and the human condition. Researchers  
in psychology and behavioural science have utilised social media sites like Twitter as a means of gathering data  
(Zimbra et al., 2018). It has also been used to the prediction of personality types and the analysis of internet user  
patterns and backgrounds. It's also intriguing to see how individuals communicate their feelings in response to  
terrible occurrences like terrorism, strong political opinions, and natural catastrophes. For example, during the  
Kenyan terror assault, Twitter served as a vital conduit for information sharing between the public, emergency  
response team, and government (Wang et al., 2012).  
Sentiment analysis is the focus of the linguistics and natural language processing area of opinion mining. In  
order to analyse and extract opinions and emotions from textual data, it assesses the degree of polarity of words  
and phrases. Numerous research projects and innovations have been undertaken by institutions or individuals  
seeking to understand public opinion on a particular subject. Furthermore, a significant amount of research has  
been done on more application-focused methodologies. Moreover, assessing Twitter conversations has been a  
fruitful field of research (Bing et al., 2012). The talk may help individuals comprehend each other's sentiments  
since it provides a plethora of discriminative information pertinent to a range of issues. Using a revolutionary  
deep learning method, the positive and negative emotions exhibited in Twitter chats were analysed. To determine  
the sentiment polarity of social media messages, it combined modules for dialogue reconstruction with emotion  
recognition. Currently assessed the dissemination of information across platforms by examining Twitter  
exchanges. created forecasting models using a latent variables-based searching approach to anticipate the  
presence of viruses based on the workload of Twitter chats (Adwan et al., 2020).  
In particular, SA is covered in great detail in this work and is divided into the following subsections: In Section  
2, the history of SA of Twitter data is shown. The issue identification of sentiment analysis is shown in Section  
www.ijltemas.in  
Page 1253  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
3. Section 4 explains the study aims, and Section 5 provides an overview of the SA approach using data from  
Twitter. Results and current developments in Twitter sentiment analysis are shown in Section 7.  
BACKGROUND  
Social studies and business have also used sentiment analysis. In order to support their industrial and commercial  
operations, corporations such as Google and Microsoft have lately developed their own sentiment analysis tools  
(Bollen et al., 2011). Sentiment analysis (SA) aims to tackle the challenge of assessing the implicit meaning  
found in tweets sent on Twitter, which is regarded as a novel topic. There are several obstacles to SA, the biggest  
of which is the message size limitation (Meng et al., 2022). Because a tweet may only be 140 characters long, it  
can be challenging to understand the meaning in that small space. Concurrently, the disorganised textual display  
on Twitter adds to the complexity. Thus, the recommended SA techniques need to address a number of issues  
(Pak et al., 2010). The typical SA operation flow is shown in Figure 1.  
Figure 1. The Operation Flow of Sentiment Analysis of Twitter Data (Giachanou et al.; 2016).  
Three basic categories may be used to categorise sentiment analysis methodologies: lexicon-based, machine  
learning-based, and hybrid techniques. Figure 2 [41] depicts the sentiment analysis taxonomy.  
Figure 2. The taxonomy of sentiment analysis (Jose et al.; 2010)  
In sentiment analysis, the classification step makes use of a classifier that has been trained using machine  
learning methods. The two main categories of this method are supervised learning and unsupervised learning.  
Table 1 provides a comprehensive publication using machine learning approaches.  
www.ijltemas.in  
Page 1254  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
Table 1. The list of machine learning-based approaches for sentiment analysis.  
Ref  
Objective and Algorithm Used  
Data Scope  
Dataset  
[11]  
Feature selection, CRF, and particle Reviews of laptops and 2014 SemEval  
swarm optimisation (PSO) restaurants  
[12]  
[13]  
[14]  
[15]  
Logistic regression model, discrete Money-related, spam- ML Respository at UCI  
PSO, and feature subset selection based, childcare, etc.  
NB, SVM, CART, Binary PSO, and Numbers typed by hand UCI  
benchmark  
feature selection  
datasets  
Choosing affective traits, multi-swarm Course Evaluation  
PSO, SVM  
Collections  
MOOC  
from  
NB, SVM, LR, feature weighting, TV, radio, medication, datasets  
taken  
from  
optimization-based  
weighted  
voting camera, etc.  
online sources  
method, Logistic regression in Bayesian  
analysis and linear discriminant  
[16]  
[17]  
SVM and binary categorization  
Review of a film  
Own  
feature weighting and Kullback-Leibler Newspaper  
divergence, SVM flexibility score  
article, Polarity  
Subjectivity  
dataset,  
dataset,  
MPQA dataset  
and movie review  
NB, SVM  
[18]  
feature selection and weighting  
movie review  
IMDb  
Table 2. The list of lexicon-based approaches proposed for sentiment analysis.  
Ref  
Objective and Algorithm Used  
Data Scope  
Dataset  
[19]  
Text classification employing finely User-generated personal story  
tuned attitude labels, taxonomy  
Dataset  
Experience  
from  
Project website  
[20]  
[21]  
[22]  
[23]  
lexicon produced by self-development, Movie review  
semantic lexicon produced by document  
discourse structure, sentiment classifier  
IMDB  
NLP, PMI-IR, and a syntax-based News information  
N/A  
comments-oriented  
analyzer  
news  
sentiment  
Comparison of supervised versus lexical Corpus of emotions  
knowledge-based approaches for  
emotion detection using SVM  
ISEAR, Emotinet  
Emotion lexicon and affect-based search Books, stories, emails, and so Corpus of enron  
www.ijltemas.in  
Page 1255  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
using crowdsourcing  
forth  
email  
[24]  
[25]  
[26]  
Rule-based classifier in an unsupervised Unlabeled SMS and Twitter SemEval  
SSA-UO system messages  
Rule-based classifier and rule-based Twitter and SMS messages  
pattern matching system  
SemEval  
Emotional  
triggers  
and  
sentiment Tweet message  
STS, OMD  
vocabulary for unsupervised sentiment  
analysis  
[27]  
[28]  
Sentiment analysis at the tweet, entity, Tweet message  
and generic sentiment lexicon levels  
OMD, HCR, STS-  
Gold  
Connotative polarity detection and a Tweet message  
vocabulary of connotations  
SemEval-2007,  
Sentiment twitter  
Problem Identification  
The problems identified by previous research (Kulkarni et al., 2022; Swathi et al., 2022) are as follows:  
• It is not always possible to identify pertinent Covid 19 tweets via retrieval.  
• It is not always possible to distinguish between positively and negatively tagged tweets.  
• Due to their limited accuracy, unrecognised tweets from social media platforms may be detected.  
Research Objectives  
The aims of the suggested work are as follows:  
• To increase accuracy so that pertinent Covid 19 positive and negative tweets are perfectly retrieved.  
• To enhance recollection throughout the retrieval procedure for perfectly relevant Covid 19 positive and  
negative tweets.  
• To increase the exactness of the train and validation accuracy in comparison to the corresponding train and  
validation loss.  
METHODOLOGY  
The algorithm of proposed model is as follows:  
Step 1. Define Network  
Keras defines a neural network as a structure composed of several layers arranged in a hierarchical manner. The  
Sequential class encompasses many levels. To begin, instantiate an object of the Sequential class. Subsequently,  
you may proceed to generate your layers and organize them in the appropriate sequence for further linkage. The  
LSTM () function refers to the recurrent layer of the LSTM, which consists of memory units. The Dense () layer  
is often positioned after LSTM layers and is used for generating prediction outputs via complete connectivity.  
The number of inputs that should be anticipated must be explicitly defined in the top layer of the network. Input  
of three-dimensional data is required, which includes samples, timesteps, and attributes.  
• Examines. The rows in your data are these.  
www.ijltemas.in  
Page 1256  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
• Steps in time. These are historical data points for a feature, like lag variables.  
• Qualities. The columns in your data are these.  
Step 2. Compile Network  
We must assemble our network once it has been defined.  
One step towards efficiency is compilation. It turns the simple layer sequence we established into a very effective  
set of matrix transformations in a format that can run on your CPU or GPU, depending on how Keras is set up.  
Step 3. Update and Optimize Weight  
Once the network is put together, it may be fit by use a training set of data to modify the weights.  
Use the effective ADAM (Adaptive Moment Estimation) optimisation technique to update the weights. For sto-  
chastic gradient descent (SGD)-based optimisation in machine learning, the Adam optimizer is an iterative op-  
timisation technique. The adaptive learning rate algorithm ADAM was created to accelerate convergence and  
increase training rates in deep neural networks.  
Step 4. Fit Network  
To fit the network, the training data must be provided. It comprises of a matrix of input patterns, X, and an array  
of matching output patterns, y.  
The network is trained using the backpropagation method, and it is optimised using the loss function and opti-  
misation strategy selected during model compilation.  
To apply the backpropagation approach, the network has to be trained for a certain number of epochs, or expo-  
sures, to the training dataset. Batches, or sets of input-output pattern pairings, may be created from each epoch.  
This indicates how many patterns the network sees before updating the weights throughout an epoch. In addition,  
it optimises efficiency by limiting the number of input patterns that are loaded into memory at once.  
Step 5. Evaluate Network  
After it has been trained, the network may be evaluated.  
Since the network has previously seen all of this data, it may be evaluated using the training set, but this won't  
provide a useful indication of how well the network works as a predictive model.  
An alternate dataset that wasn't utilised for testing might be used to evaluate the network's performance. In the  
case of unobserved data, this will provide an estimate of the network's prediction performance.  
In addition to any other metrics (such classification accuracy) that were provided during model compilation, the  
model evaluates the loss over all test patterns. A set of evaluation metrics is returned.  
Step 6. Make Predictions  
If we are happy with our fit model's performance, we could utilise it to make predictions on new data. To do  
this, just call the model's predict() method with a range of new input patterns.  
Experiment and Result  
This section outlines the precise procedures of the experiment after outlining various presumptions and  
constraints. The following are the presumptions made in this work:  
www.ijltemas.in  
Page 1257  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
(1) The training and testing data come from a single dataset, and our study focuses on sentiment analysis of  
COVID-19 tweets. When using the ant dataset for experimentation, for instance, the training set is chosen and  
the test set is created from the remaining portion of the dataset.  
(2) The trained model favours the negative tweets classes during the trials due to a limited number of faulty  
classes. Thus, before training the model, class imbalance is applied to the whole dataset.  
(3) An optimisation algorithm called ADAM (Adaptive Moment Estimation) is employed to more accurately  
evaluate the method performance.  
The correctness of the sentiment analysis paradigm suggested in this study may be confirmed under the  
aforementioned presumptions. The particular protocol for the experiment is as follows:  
Step 1: The software's class dependency is extracted using the code analysis tool, and a CSV file is subsequently  
created.  
Step 2: The covid 19 tweets dataset is used to extract the labelled nodes and feature metrics for each node.  
Step 3: To address the imbalance in data classes, the LSTM-ADAM technique is used.  
Figure 3. Calculation of confusion matrix, precision, recall, F1-Score and accuracy among different models and  
LSTM-ADAM (Proposed Model)  
Table 4: Estimation of Sentiment Parameters using LSTM-ADAM (Proposed Model)  
Sentiment Parameter  
Positive  
Count  
11422  
9917  
7713  
6624  
5481  
Negative  
Neutral  
Extremely Positive  
Extremely Negative  
Figure 4. Graphical Analysis of Sentiment Parameters using LSTM-ADAM (Proposed Model)  
www.ijltemas.in  
Page 1258  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
Figure 5. Spherical Analysis of Sentiment Proportions LSTM-ADAM (Proposed Model)  
Table 5. Estimation of Precision, Recall and F1-Score among different models and LSTM-ADAM (Proposed  
Model) for Negative Tweets  
Models  
Precision  
0.55  
Recall  
0.40  
0.68  
0.75  
0.6  
F1-Score  
0.46  
KNN  
Decision Tree  
SVM  
0.69  
0.68  
0.77  
0.76  
XG Boost  
0.7  
0.79  
LSTM-ADAM  
Model)  
(Proposed 0.79  
0.79  
0.79  
1
0.8  
0.6  
0.4  
0.2  
0
KNN  
Decision Tree  
SVM  
XG Boost  
LSTM-ADAM  
(Proposed  
Model)  
Figure 6. Graphical Analysis of Precision among different models and LSTM-ADAM (Proposed Model) for  
Negative Tweets  
The above graph show that the proposed model gives better precision for negative tweets as compare than other  
models. The precision of LSTM-ADAM (Proposed Model) is improve by 0.02 as compare than SVM prediction  
model.  
www.ijltemas.in  
Page 1259  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
1
0.8  
0.6  
0.4  
0.2  
0
KNN  
Decision Tree  
SVM  
XG Boost  
LSTM-ADAM  
(Proposed  
Model)  
Figure 7. Graphical Analysis of Recall among different models and LSTM-ADAM (Proposed Model) for  
Negative Tweets  
The above graph show that the proposed model gives better recall for negative tweets as compare than other  
models. The recall of LSTM-ADAM (Proposed Model) is improve by 0.04 as compare than SVM prediction  
model.  
0.9  
0.8  
0.7  
0.6  
0.5  
0.4  
0.3  
0.2  
0.1  
0
KNN  
Decision Tree  
SVM  
XG Boost  
LSTM-ADAM  
(Proposed  
Model)  
Figure 8. Graphical Analysis of F1-Score among different models and LSTM-ADAM (Proposed Model) for  
Negative Tweets  
The above graph show that the proposed model gives better F1-Score for negative tweets as compare than other  
models. The F1-Score of LSTM-ADAM (Proposed Model) is similar to XG Boost Classifier model.  
Table 6. Estimation of Precision, Recall and F1-Score among different models and LSTM-ADAM (Proposed  
Model) for Neutral Tweets  
Models  
Precision  
0.25  
Recall  
0.72  
0.67  
0.66  
0.65  
0.68  
F1-Score  
0.37  
KNN  
Decision Tree  
0.62  
0.64  
SVM  
0.65  
0.66  
XG Boost  
0.75  
0.68  
LSTM-ADAM (Proposed Model)  
0.68  
0.68  
www.ijltemas.in  
Page 1260  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
0.8  
0.7  
0.6  
0.5  
0.4  
0.3  
0.2  
0.1  
0
KNN  
Decision Tree  
SVM  
XG Boost  
LSTM-ADAM  
(Proposed  
Model)  
Figure 9. Graphical Analysis of Precision among different models and LSTM-ADAM (Proposed Model) for  
Neutral Tweets  
The above graph show that the XG-Boost model gives better precision for neutral tweets as compare than other  
models.  
0.74  
0.72  
0.7  
0.68  
0.66  
0.64  
0.62  
0.6  
KNN  
Decision Tree  
SVM  
XG Boost  
LSTM-ADAM  
(Proposed  
Model)  
Figure 10. Graphical Analysis of Recall among different models and LSTM-ADAM (Proposed Model) for  
Neutral Tweets  
The above graph show that the KNN model gives better recall for neutral tweets as compare than other models.  
0.8  
0.7  
0.6  
0.5  
0.4  
0.3  
0.2  
0.1  
0
KNN  
Decision Tree  
SVM  
XG Boost  
LSTM-ADAM  
(Proposed  
Model)  
Figure 11. Graphical Analysis of F1-Score among different models and LSTM-ADAM (Proposed Model) for  
Neutral Tweets  
www.ijltemas.in  
Page 1261  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
The above graph show that the proposed model gives better F1-Score for neutral tweets as compare than other  
models. The F1-Score of LSTM-ADAM (Proposed Model) is similar to XG Boost Classifier model.  
Table 7. Estimation of Precision, Recall and F1-Score among different models and LSTM-ADAM (Proposed  
Model) for Positive Tweets  
Models  
Precision  
0.69  
Recall  
0.29  
0.74  
0.81  
0.86  
0.83  
F1-Score  
0.41  
KNN  
Decision Tree  
0.75  
0.75  
SVM  
0.8  
0.8  
XG Boost  
0.55  
0.67  
LSTM-ADAM (Proposed Model)  
0.82  
0.82  
0.9  
0.8  
0.7  
0.6  
0.5  
0.4  
0.3  
0.2  
0.1  
0
KNN  
Decision Tree  
SVM  
XG Boost  
LSTM-ADAM  
(Proposed  
Model)  
Figure 12. Graphical Analysis of Precision among different models and LSTM-ADAM (Proposed Model) for  
Positive Tweets  
The above graph show that the proposed model gives better precision for positive tweets as compare than other  
models. The precision of LSTM-ADAM (Proposed Model) is better than SVM Classifier model.  
1
0.9  
0.8  
0.7  
0.6  
0.5  
0.4  
0.3  
0.2  
0.1  
0
KNN  
Decision Tree  
SVM  
XG Boost  
LSTM-ADAM  
(Proposed  
Model)  
Figure 13. Graphical Analysis of Recall among different models and LSTM-ADAM (Proposed Model) for  
Positive Tweets  
www.ijltemas.in  
Page 1262  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
The above graph show that the XG-Boost model gives better recall for positive tweets as compare than other  
models.  
0.9  
0.8  
0.7  
0.6  
0.5  
0.4  
0.3  
0.2  
0.1  
0
KNN  
Decision Tree  
SVM  
XG Boost  
LSTM-ADAM  
(Proposed  
Model)  
Figure 14. Graphical Analysis of F1-Score among different models and LSTM-ADAM (Proposed Model) for  
Positive Tweets  
The above graph show that the proposed model gives better F1-Score for positive tweets as compare than other  
models. The F1-Score of LSTM-ADAM (Proposed Model) is better than SVM model.  
Table 8. Estimation of Accuracy among different models and LSTM-ADAM (Proposed Model) for Positive  
Tweets  
Models  
Accuracy  
0.41  
KNN  
Decision Tree  
0.7  
SVM  
0.76  
XG Boost  
0.6  
LSTM-ADAM (Proposed Model)  
0.78  
0.9  
0.8  
0.7  
0.6  
0.5  
0.4  
0.3  
0.2  
0.1  
0
KNN  
Decision Tree  
SVM  
XG Boost  
LSTM-ADAM  
(Proposed  
Model)  
Figure 15. Graphical Analysis of Accuracy among different models and LSTM-ADAM (Proposed Model) for  
Covid-19 Tweets  
www.ijltemas.in  
Page 1263  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
The above graph show that the proposed model gives better accuracy for covid 19 tweets as compare than other  
models. The accuracy of LSTM-ADAM (Proposed Model) is better than SVM Classifier model.  
CONCLUSİONS  
Researchers should focus on this topic because of the significant rise of digital text data on servers and libraries.  
In light of this, study has concentrated on the problem of content sentiment identification. Although a great deal  
of research has previously been done in this area, it has only focused on semi-automated sentiment  
categorization. The suggested improvement will improve the work's sentiment detection effectiveness across all  
assessment metrics. The whole model uses dictionaries to preprocess the input before pulling patterns and  
keywords from it. The learning model's accuracy increases with the decrease of the input feature set, where a  
geo-inspired technique yields a high-quality vector set. Therefore, it is anticipated that the study will provide a  
model that identifies the sentiment of tweet content where the material does not need any structure and the  
algorithm's execution time is also quite short.  
The following are the thesis work's conclusions:  
• Lemmatizing and stemming text material from tweets to achieve ideal sentiment content categorization.  
• The suggested model supports any manual participation for sentiment recognition; • The LSTM model may  
accept text information in any format;  
• The ADAM approach is used to optimise the learning model by decreasing the feature vector. While the SVM  
methodology has an accuracy of around 76%, the suggested method (LSTM-ADAM) has an accuracy of 78%.  
REFERENCES  
1. Zimbra, D.; Abbasi, A.; Zeng, D.; Chen, H. The State-of-the-Art in Twitter Sentiment Analysis: A  
Review and Benchmark Evaluation. ACM Trans. Manag. Inf. Syst. (2018)  
2. Bollen, J.; Mao, H.; Zeng, X. Twitter Mood Predicts the Stock Market. J. Comput. Sci. (2011).  
3. Wang, H.; Can, D.; Kazemzadeh, A.; Bar, F.; Narayanan, S. A System for Real-time Twitter Sentiment  
Analysis of 2012 U.S. Presidential Election Cycle. In Proceedings of the ACL 2012 System  
Demonstrations, Jeju Island, Korea, 10 July (2012).  
4. Meng, F.; Xiao, X.; Wang, J. Rating the crisis of online public opinion using a multi-level index system.  
Int. Arab. J. Inf. Techn. (2022)  
5. Bing, L.; Lei, Z. A survey of opinion mining and sentiment analysis. In Mining Text Data, 1st ed.; Charu,  
C.A., ChengXiang, Z., Eds.; Springer: New York, NY, USA, (2012).  
6. Adwan, O.; Al-Tawil, M.; Huneiti, A.; Shahin, R.; Zayed, A.A.; Al-Dibsi, R. Twitter sentiment analysis  
approaches: A survey. Int. J. Emerg. Technol. (2020)  
7. Giachanou, A.; Crestani, F. Like it or not: A survey of Twitter sentiment analysis methods. ACM  
Comput. Surv. (CSUR) (2016)  
8. Pak, A.; Paroubek, P. Twitter as a corpus for sentiment analysis and opinion mining. In Proceedings of  
the Seventh International Conference on Language Resources and Evaluation, Valletta, Malta, 17–23  
May 2010. Appl. Sci. (2022)  
9. Swathi, T.; Kasiviswanath, N.; Rao, A.A. An optimal deep learning-based LSTM for stock price  
prediction using twitter sentiment analysis. Appl. Intell. (2022)  
10. Jose, A.K.; Bhatia, N.; Krishna, S. Twitter Sentiment Analysis; Seminar Report; National Institute of  
Technology Calicut: Kozhikode, India, (2010).  
11. Gupta, D.K.; Reddy, K.S.; Ekbal, A. Pso-asent: Feature selection using particle swarm optimization for  
aspect based sentiment analysis. In Proceedings of the International Conference on Applications of  
Natural Language to Information Systems, Passau, Germany, 17 June (2015).  
12. Unler, A.; Murat, A. A discrete particle swarm optimization method for feature selection in binary  
classification problems. Eur. J. Oper. Res. (2010)  
13. Shang, L.; Zhou, Z.; Liu, X. Particle swarm optimization-based feature selection in sentiment  
www.ijltemas.in  
Page 1264  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
classification. Soft Comput. (2016)  
14. Liu, Z.; Liu, S.; Liu, L.; Sun, J.; Peng, X.; Wang, T. Sentiment recognition of online course reviews using  
multi-swarm optimization based selected features. Neurocomputing (2016)  
15. Onan, A.; Koruko ˘glu, S.; Bulut, H. A multiobjective weighted voting ensemble classifier based on  
differential evolution algorithm for text sentiment classification. Expert Syst. Appl. (2016)  
16. Basari, A.S.H.; Hussin, B.; Ananta, I.G.P.; Zeniarja, J. Opinion mining of movie review using hybrid  
method of support vector machine and particle swarm optimization. Procedia Eng. (2013)  
17. Kummer, O.; Savoy, J. Feature Weighting Strategies in Sentiment Analysis. In Proceedings of the First  
International Workshop on Sentiment Discovery from Affective Data, Bristol, UK, 24–28 September  
(2012)  
18. O’Keefe, T.; Koprinska, I. Feature selection and weighting methods in sentiment analysis. In Proceedings  
of the 14th Australasian Document Computing Symposium, Sydney, NSW, Australia, 4 December  
(2009)  
19. Neviarouskaya, A.; Prendinger, H.; Ishizuka, M. Recognition of affect, judgment, and appreciation in  
text. In Proceedings of the 23rd International Conference on Computational Linguistics, Beijing, China,  
23–27 August (2010).  
20. Heerschop, B.; Goossen, F.; Hogenboom, A.; Frasincar, F.; Kaymak, U.; de Jong, F. Polarity analysis of  
texts using discourse structure. In Proceedings of the 20th ACM International Conference on Information  
and Knowledge Management, Glasgow, UK, 24–28 October (2011).  
21. Moreo, A.; Romero, M.; Castro, J.L.; Zurita, J.M. Lexicon-based comments-oriented news sentiment  
analyzer system. Expert Syst. Appl. (2012)  
22. Balahur, A.; Hermida, J.M.; Montoyo, A. Detecting implicit expressions of emotion in text: A  
comparative analysis. Decis. Support Syst. (2012)  
23. Mohammad, S.M. From once upon a time to happily ever after: Tracking emotions in mail and books.  
Decis. Support Syst. (2012)  
24. Ortega, R.; Fonseca, A.; Montoyo, A. SSA-UO: Unsupervised Twitter sentiment analysis. In Proceedings  
of the Second Joint Conference on Lexical and Computational Semantics, Atlanta, GA, USA, 13–14 June  
(2013)  
25. Reckman, H.; Baird, C.; Crawford, J.; Crowell, R.; Micciulla, L.; Sethi, S.; Veress, F. teragram: Rule-  
based detection of sentiment phrases using sas sentiment analysis. In Proceedings of the Second Joint  
Conference on Lexical and Computational Semantics, Volume 2: Seventh International Workshop on  
Semantic Evaluation, Atlanta, GA, USA, 14–15 June 2013; Association for Computational Linguistic:  
Stroudsburg, PE, USA, (2013)  
26. Hu, X.; Tang, J.; Gao, H.; Liu, H. Unsupervised sentiment analysis with emotional signals. In  
Proceedings of the 22nd International Conference on World Wide Web, Rio de Janeiro, Brazil, 13 May  
(2013)  
27. Saif, H.; He, Y.; Fernandez, M.; Alani, H. Contextual semantics for sentiment analysis of Twitter. Inf.  
Process. Manag. (2016)  
28. Feng, S.; Bose, R.; Choi, Y. Learning general connotation of words using graph-based algorithms. In  
Proceedings of the Conference on Empirical Methods in Natural Language Processing, Edinburgh, UK,  
27–31 July (2011)  
29. Chen, L.S.; Liu, C.H.; Chiu, H.J. A neural network based approach for sentiment classification in the  
blogosphere. J. Informetr. (2011)  
30. Xia, R.; Zong, C.; Li, S. Ensemble of feature sets and classification algorithms for sentiment  
classification. Inform. Sci. (2011)  
31. Abdul-Mageed, M.; Diab, M.; Kübler, S. SAMAR: Subjectivity and sentiment analysis for Arabic social  
media. Comput. Speech Lang (2014)  
32. Abbasi, A.; France, S.; Zhang, Z.; Chen, H. Selecting attributes for sentiment classification using feature  
relation networks. IEEE Trans. Knowl. Data Eng. (2011)  
33. Prabowo, R.; Thelwall, M. Sentiment analysis: A combined approach. J. Informetr. (2009)  
34. Zhang, L.; Ghosh, R.; Dekhil, M.; Hsu, M.; Liu, B. Combining Lexicon-Based and Learning-Based  
Methods for Twitter Sentiment Analysis; Technical Report HPL-2011-89; Hewlett-Packard  
Development Company: Palo Alto, CA, USA, (2011)  
www.ijltemas.in  
Page 1265  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
35. Ghiassi, M.; Skinner, J.; Zimbra, D. Twitter brand sentiment analysis: A hybrid system using n-gram  
analysis and dynamic artificial neural network. Expert Syst. Appl. (2013)  
36. Khuc, V.N.; Shivade, C.; Ramnath, R.; Ramanathan, J. Towards building large-scale distributed systems  
for Twitter sentiment analysis. In Proceedings of the 27th Annual ACM Symposium on Applied  
Computing, Trento, Italy, 26–30 March (2012)  
37. Khan, F.H.; Bashir, S.; Qamar, U. TOM, Twitter opinion mining framework using hybrid classification  
scheme. Decis. Support Syst. (2014)  
www.ijltemas.in  
Page 1266