IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 13, No. 3, September 2024, pp. 3395~3403

ISSN: 2252-8938, DOI: 10.11591/ijai.v13.i3.pp3395-3403  3395

Elevating sentiment analysis with deep convolutional neural

network model facial expression insights

Pradnya Samit Mehta1, Renuka Wankhade2, Dev Pankaj Chhabada3, Dhimahi Patel3,
Anirudh Kailas Gote3, Anuradha Yenkikar4, Poorva Agrawal5, Gagandeep Kaur5
1
Department of Computer Engineering, Vishwakarma Institute of Information Technology, Pune, India
2
Department of Artificial Intelligence and Machine Learning, Vishwakarma Institute of Information Technology, Pune, India
3
Department of Artificial Intelligence and Data Science, Vishwakarma Institute of Information Technology, Pune, India
4
Department of Computer Science-Artificial Intelligence, Vishwakarma Institute of Information Technology, Pune, India
5
Symbiosis Institute of Technology, Nagpur Campus, Symbiosis International (Deemed University), Pune, India

Article Info ABSTRACT

Article history: In today's data-driven world, the ability to analyze emotional responses is
essential. The pressing necessity that drives this study is to revolutionize the
Received Dec 21, 2023 field of sentiment analysis by extracting the hidden information from people's
Revised Mar 15, 2024 facial expressions. It examines people's preferences, worries, and pleasure,
Accepted Mar 23, 2024 revealing their views on many topics. Beyond text-based sentiment analysis,
this research adds facial expression-based sentiment analysis into existing
systems for tailored recommendations and mental health monitoring. The
Keywords: system emphasizes visual stimuli's emotional influence to improve decision-
making, content adaptability, and user experiences. The implementation
Convolutional neural network involves transfer learning with the pre-trained VGG-16 model, which
Emotion enhances ability to discern intricate emotional cues from facial expressions.
Facial expression-based analysis Convolutional neural network (CNN) and contextual analysis allow the model
Machine learning to understand users' emotions and provide insights into their thoughts,
Sentiment analysis feelings, and behaviors. To improve emotion recognition reliability and
reactivity, this study examines random forest, support vector machine (SVM),
and CNN methodologies. The visual geometry group (VGG-16) CNN model
outperforms over SVM and random forest classifiers with accuracy of 95%.
This study highlights facial expression-based sentiment analysis.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Pradnya Samit Mehta
Department of Computer Engineering, Vishwakarma Institute of Information Technology
Pune, India
Email: [email protected]

1. INTRODUCTION
The emergence of the internet and other digital technologies has brought about a new era of
communication that heavily relies on facial expressions. Whether it's for entertainment, education, or
interpersonal interactions, visual media plays a significant role in everything to do. With the proliferation of
facial expression-sharing apps, social media platforms, and electronic communication, it has become essential
to accurately interpret the emotional content conveyed through visual imagery. The impact of emotions on well-
being and decision-making is evident in various aspects of our lives, including our physical and mental health.
The ability to capture and understand emotions has numerous potential applications, ranging from improving
mental healthcare to developing compassionate artificial intelligence (AI)-driven interactions. Traditionally,
sentiment analysis focused on analyzing textual data, such as words and phrases, to gauge emotional tones.
However, gaining a deeper understanding of human emotions requires more than just textual analysis.

Journal homepage: http://ijai.iaescore.com

3396  ISSN: 2252-8938

The ability for robots to understand and interpret human emotions in multimedia environments is a
major goal of the field of sentiment analysis. At its foundation, it is the process of identifying and classifying
people's feelings, thoughts, and attitudes regarding particular persons, products, events, or topics [1]. Despite
the advancements in sentiment analysis, accurately deciphering emotions from images continues to be a major
hurdle. The intricate complexity of human facial expressions, along with the variations in lighting conditions
and angles, significantly hinder the effectiveness of existing methodologies. Consequently, there is a pressing
need for a robust solution that can discern emotions with unparalleled accuracy and reliability. Sentiment
analysis has come a long way since its early days. Research in sentiment analysis has advanced significantly
with the growth of new venues, such as blogs and social media, where individuals regularly share their ideas
[2]. Social media's pervasiveness has brought the world's population closer together, allowing for the
uninhibited exchange of ideas [3]. By utilizing the advanced capabilities of visual geometry group (VGG-16),
cutting-edge convolutional neural network (CNN) architecture renowned for its exceptional image recognition
and feature extraction abilities, aimed to delve further into the intricacies of facial expression analysis. The
driving force behind our study originates from the recognition of facial expressions as a universal language of
emotions that surpasses linguistic and cultural barriers. Through harnessing the valuable insights provided by
facial expressions, our objective is to improve the accuracy and granularity of sentiment analysis, ultimately
leading to a deeper comprehension of human emotions.
Every aspect of sentiment analysis how it is conducted, the data it uses, and the questions it asks—is
evolving as a result of technological developments. In the early 2000s, researchers began using simple
techniques like keyword-based analysis to determine if a piece of writing should be classified as positive,
negative, or neutral. The development of natural language processing (NLP) tools, however, has ushered in a
new era of sentiment analysis, marked by more refined approaches. Accuracy was improved by employing
machine learning methods and features like n-grams and syntactic structures. Due to the immediacy and
informality of user-generated material, the rise of social media platforms in the late aughts presented new
issues. Improvements in sentiment analysis can be attributed in large part to the introduction of cutting-edge
methodologies, such as lexicon-based systems that calculate sentiment scores and classify evaluations into
positive or negative categories [4]. Support vector machines (SVM), naive bayes, and neural networks have
become popular methods for sentiment analysis since the advent of machine learning and deep learning [5].
While aspect-based sentiment analysis made it possible to pinpoint opinions about individual items in text,
domain adaption techniques worked to enhance sentiment analysis in general [6]. Sentiment analysis in
multiple languages was a problem that multilingual sentiment analysis set out to solve [7]. It is still evolving
as a method for gaining understanding from a wide variety of textual data in numerous fields and languages.
In light of these advancements, the proposed method may benefit from this study to provide a detailed
comparison of the many algorithms such as SVM, VGG-16 CNN, and random forest for predicting the
sentiment analysis. The proposed work also recommends creating a facial expression-based sentiment analysis
algorithm for properly gauging and reacting to people' emotions and moods with digital world. In fields like
mental health, where understanding and managing emotional states is crucial to individual well-being, this
approach has the potential to dramatically boost our ability to give effective assistance and therapies. The key
goals of this study are to develop a facial expression-based sentiment analysis model that is capable of
recognizing and categorizing user emotions based on input frames of facial expression expressions. The
integration of computer vision and sentiment analysis is underscored, recognizing the harmonious connection
between these disciplines in offering valuable insights into human sentiment and behavior. In this study, a
proposed method will be utilized to examine the sentiment analysis VGG-16 CNN, SVM, random forest
algorithms. These algorithms will be evaluated and compared using various metrics including accuracy,
precision, recall, and F1-score, to determine their relative rankings w.r.t accuracy and granularity of sentiment
analysis.

2. LITERATURE SURVEY
The purpose of this literature review is to present a summary of current studies conducted in the field
of sentiment analysis, with an eye toward how these studies relate to our own. The information gleaned from
this survey will be invaluable in determining the course of our future studies. The authors suggest a lexical
knowledge-based extraction strategy to achieve this comprehension from video transcriptions. They employ
SenticNet to parse MuSe-CAR data for linguistic concepts and refine a number of feature types. They used the
MuSe-automobile dataset to do research on sentiment analysis in the context of YouTube facial expression
automobile evaluations. They used the Senticnet framework and a machine learning strategy based on SVM,
achieving a combined metric of 66.16% on the test set and 56.18% on the development set. Surprisingly, their
SVM-based model performed better than long short-term memory (LSTMs) units, popular neural network
architecture, by more than 30%. The best model they present increases the linguistic baseline from the

Int J Artif Intell, Vol. 13, No. 3, September 2024: 3395-3403

Int J Artif Intell ISSN: 2252-8938  3397

MuSe-Topic 2020 sub-challenge by roughly 3% (absolute) in the prediction of valence on the standard
challenge measure [8].
In a different piece of research, author suggested a deep neural network (DNN)-based model for
sentiment analysis. In order to train their AI model, they used facial expression data from the YouTube channel
‘Joice Hasselmann’, from which they extracted opinion elements from raw text comments and then sorted them
into discrete categories along a continuous scale from 1 to +1. The 84% accuracy achieved by their 6-layer
DNN on the test set is indicative of the model's robust predictive capacity. To possibly improve results, the
authors advised additional data preprocessing [9].
This study investigates the potential of gated recurrent units (GRUs), recurrent neural network
architecture for assessing emotional valence in textual input. It suggests an extensive review of the use of deep
learning methods for sentiment categorization and analysis. The Amazon review dataset was used to create
several baseline models, such as LSTMs, GRUs, bidirectional long short-term memory (Bi-LSTM), and
bidirectional gated recurrent unit (Bi-GRU). In contrast to LSTM, GRU were found to be more effective and
to have shorter training times. Accuracy of 71.19% was achieved by the Bi-GRU, with remarkable precision,
recall, and F1 scores of 71.3% each [10].
CNN were recently studied and highlighted its usefulness in sentiment analysis and how they can
capture hierarchical patterns in textual data. CNNs used for analyzing English and Chinese text and found that
the former had a 92% accuracy rate and the latter had an 87% accuracy rate. When applied to Hindi movie
reviews, it showed exceptional adaptability across languages, obtaining 95.4% accuracy [11]. Authors
presented the word2vec and CNN framework for the purpose of sentiment analysis; they entered this data into
a CNN architecture using normalization and dropout techniques. On a dataset consisting of critical assessments
of films, their model outperformed RNNs and MV-RNNs with a test accuracy of 45.4%. Their findings
highlighted the significance of pre-training and fine-tuning for accuracy in CNNs [12], [13]. Using a CNN,
researcher was able to classify images with an accuracy of 72.40% during validation. The accuracy of the
model was dramatically raised to 95.40% with the use of image enhancement techniques and transfer learning
using the VGG-16 pretrained model. This proved the value of using pretrained models like VGG-16 for
improved image categorization [14]. Authors suggested a CNN-based approach that merged text and image
data in tweets for sentiment analysis, with accuracy rates of 78% and 79.6% on various datasets, respectively,
exceeding conventional techniques. By highlighting the capability to grasp intricate correlations between text
and images, their study bridged the gap between CNNs' triumphs in text and image sentiment analysis [15]. A
study on picture classification in which they used a number of different classification models in conjunction
with deep features retrieved with the VGG19 CNN. Compared to other classifiers and approaches, random
forest's combined feature accuracy of 93.73% is striking. A combination strategy for picture categorization was
proven beneficial in this study [16]. The applicability of various deep learning models for sentiment analysis
of images, including DNNs, CNN, region-based convolutional neural network (R-CNNs), and Fast R-CNNs
investigated. According to their research, CNNs are more productive and precise than competing networks
[17]. SVM and CNN were compared for image classification. While SVM performed well on a tiny dataset,
CNN performed much better, with an astounding 93.57% accuracy on the same dataset [18]. The "Viola Jones
face detection" technique was used for face localization in a work on facial expression identification and
sentiment analysis. Zernike moments, linear binary patterns, and the discrete cosine transform were among the
feature extraction techniques and classifiers employed. On the JAFFE database, their algorithm recognized
90.14% of the images [19]. The CNN was used to extract features from the images, and the VGG was used to
analyze the emotions conveyed in the photo. The classification of buccal squamous cell carcinoma with VGG16
achieved an accuracy of 83.33 percent on training data [20]. This literature review covers a wide range of
studies and methods within the field of sentiment analysis, with a focus on how deep learning methods, CNN,
and the combination of text and image data have contributed to the development of sentiment analysis in a
number of contexts and languages.

3. PROPOSED METHODOLOGY
Recommended strategy is built on utilizing the powerful architecture of CNNs to create a facial
expression-based sentiment analysis model. The goal is to automatically recognize and understand the range
of emotions conveyed by facial expressions using CNN. Autism spectrum disorder, anorexia, and other forms
of teenage mental disease can be predicted using sentiment analysis applied to current data gathered from social
networking sites [21]. In order to effectively distinguish and categorize a wide range of emotions, the model
will be trained using a dataset of facial expressions.

3.1. System design

The suggested model, as seen in Figure 1, starts with data collection—in this example, pictures of
faces with expressions that represent different emotions. Facial expression recognition and audio processing
Elevating sentiment analysis with deep convolutional neural network model facial … (Pradnya Samit Mehta)
3398  ISSN: 2252-8938

are two strategies used in the feature extraction process to retrieve relevant data. The VGG-16 CNN algorithm
is divided into number of sub processes like pooling layer, NLP, and regularization. The model is detecting
mental health as a result with the proposed system.

Figure 1. System architecture

3.1.1. Dataset
The first step in the proposed model as shown in Figure 1 is to collect data, in this case images of face
expressions conveying various emotions. The process of feature extraction makes use of audio processing and
facial expression detection to glean useful information. The dataset is gathered from Kaggle and it consists of
19 classes which contain 8 different types of emotions of different people. Images are categorized based on
different kind of facial expression expressions like disgust, fear, happy, neutral, sad, surprised, anger and
contempt. A dataset which is used for sentiment analysis purposes is likely to contain a variety of attributes or
features related to sentiment, such as emotion-based images and their corresponding textual descriptions or
annotations [22].
Data pre-processing encompasses steps like data normalization and data enhancement. TensorFlow's
ImageDataGenerator to take care of pre-processing and data preparation for testing and training. The feature
set is improved through feature selection and engineering, and the VGG-16 CNN model is used for sentiment
analysis. The 'ImageDataGenerator' tool, found in the TensorFlow and Keras libraries, is frequently used for
enhancing and pre-processing image data before to use in deep learning tasks. Using a CNN architecture that
is good at capturing spatial data is crucial when classifying images. The primary splitting function is provided
by the 'validation_split' argument. This information set is composed of 80% training data and 20% validation
data. The next steps include training, validation, and hyper-parameter adjustment. Evaluation metrics including
accuracy, precision, recall, and F1-score matrices are used to assess the performance of a model. Studying
facial expressions can help predict emotions in real time, and response processes enable sympathetic dialogue.
After that, future study areas in mental healthcare and content customization can be identified and incorporated
into existing systems.

3.2. System implementation

The VGG-16 CNN as shown in Figure 2 has been trained on a massive and varied picture dataset,
granting it the capacity to autonomously extract meaningful features from images using a series of
convolutional and pooling processes. This model with pre-trained ImageNet weights using VGG-16. The
proposed method has been opted to add more classification layers to it [23]. After developing a unique model,
it is evaluated and refined with the help of a specific dataset. The CNN layers are discussed in following section.

Figure 2. Emotion recognition using VGG16-CNN

Int J Artif Intell, Vol. 13, No. 3, September 2024: 3395-3403

Int J Artif Intell ISSN: 2252-8938  3399

3.3. Convolution layer

The convolutional layer is designed to acquire a range of features from the input, including edges,
textures, patterns, and more intricate structures. These features are subsequently transmitted to succeeding
layers for additional analysis. Convolutional layers offer significant advantages, such as the capacity to grasp
spatial hierarchies within data, decrease the parameter count in contrast to fully connected layers, and facilitate
the detection of translation-invariant features.

3.4. Pooling layer

Pooling layers, which function similarly to summarizers, are frequently used after this scanning
procedure. They aid in distillation, allowing us to store only the most relevant details and dismiss the rest. This
facilitates the model's ability to zero in on the most relevant portions of the data [24]. The convoluted image is
calculated as shown in (1). The convolution process computes the output feature map Y from the input feature
map (tensor) X and the convolutional kernel (tensor) W by element-wise multiplying and summing. Y is
element-wise subjected to the ReLU activation function, as indicated in (2).

𝑠(𝑖, 𝑗) = ∑𝑚 ∑𝑛 ⬚ (𝐼(𝑚, 𝑛)𝑘(𝑖 − 𝑚, 𝑗 − 𝑛) (1)

𝑌 = 𝑚𝑎𝑥( 0, 𝑌) (2)

The global average pooling layer is used to retain essential information while decreasing the spatial
dimensions of the feature maps produced by the convolutional layers. Many models use it instead of regular
fully linked layers to cut down on the number of parameters and the risk of overfitting. Given the feature map
X from the previous layer, the global average pooling computes the average value along each channel (feature
map) is derived from (3).

𝑌𝑘 = 1/𝐻 + 𝑊 ∑𝐻 𝑊
𝑖=1 ⬚ ∑𝑗=1 ⬚ 𝑋𝑖𝑗𝑘 (3)

Here, H and W are the height and width of the feature map.

3.5. Dense layer

Dense layers, also known as fully connected layers, are a fundamental component of neural networks
used for classification and regression tasks. They serve to connect all neurons in one layer to all neurons in the
subsequent layer. In the context of classification, the output of a dense layer often represents the scores or
probabilities for each class. Given an input vector X and weight matrix W, the dense layer computes the output
vector Y, b is the bias vector in (4).

𝑌 =𝑋∗𝑊+𝑏 (4)

3.6. Activation function

The ReLU activation function is applied element-wise to Y: Y=max (0, Y) (for hidden layers). The
softmax activation function is applied to the output layer to produce class probabilities. These equations
represent the operations performed by the convolutional layers, global average pooling layer, and dense layers
in the code you provided. The code uses TensorFlow, which handles the computation of gradients and updates
to weights and biases during backpropagation automatically [25].

4. RESULTS AND DISCUSSION

VGG-16 CNN, SVM [26], and random forest were the three methods of sentiment analysis that were
compared by researchers and contrast in Table 1. Accuracy is 95%, 74%, and 79% respectively. VGG-16 CNN
is outperforming among all w.r.t precision, recall, F1-score and accuracy.
The model's accuracy is quantified by the percentage of training instances that were properly labeled.
A high accuracy during training suggests that the model is successfully adapting to fit the provided data [27].
The correctness of the model's validation is evaluated by how well it performs on a different dataset known as
the validation dataset. Instead of being used for training, this dataset is just used to evaluate the model's ability
to generalize to novel data. If the model has good performance on data it hasn't encountered during training, it
means the validation accuracy is high. In a perfect world, the model's learning and generalization abilities
would be reflected in high and similar training and validation accuracy.

Elevating sentiment analysis with deep convolutional neural network model facial … (Pradnya Samit Mehta)
3400  ISSN: 2252-8938

Table 1. The performance comparative of VGG-16 CNN, SVM, and random forest
Method Precision Recall F1-score Accuracy
VGG-16 CNN 0.95 0.93 0.94 0.95
SVM 0.68 0.90 0.67 0.74
Random forest 0.72 0.79 0.74 0.79

The accuracy of the model's predictions is evaluated during training by comparing them to the true
labels. It is a common metric used to assess the degree of discrepancy between expected and observed
outcomes. Training aims to reduce this loss as much as possible. Similar to training loss, but determined using
the validation dataset, is validation loss. It evaluates how closely the model's predictions match the validation
data's actual labels. Training loss should go down steadily in a good scenario, showing that the model is
improving. The Figures 3 to 5 depicts the confusion matrix, graphical analysis of training and validation
accuracy as well as loss with respect to the proposed model.

Figure 3. Confusion matrix

Figure 4. Training and validation loss Figure 5. Training and validation accuracy

As was indicated in Figure 5, the accuracy reaches around 98%, which is in line with what is to be
expected from a high-quality model. The loss also goes down and down until it's practically gone, which is a

Int J Artif Intell, Vol. 13, No. 3, September 2024: 3395-3403

Int J Artif Intell ISSN: 2252-8938  3401

good thing and in line with our predictions. In addition, it can be shown that the model improves with more
epochs, leading to better accuracy and less loss.

5. CONCLUSION
In this research, researchers took benefit of transfer learning by using CNN architecture, VGG-16
CNN, which had already been trained on the massive ImageNet dataset in order to improve our ability to
decode the emotions conveyed by facial expressions. CNN from the field of computer vision play a central role
in our novel approach, which has its origins in facial expression-based sentiment analysis. This innovative
method exemplifies the vast potential of interdisciplinary cooperation in the field of sentiment analysis.
Proposed system extracts relevant features from facial expression frames using pre-trained deep learning
models like VGG-16 CNN which demonstrates the power of transfer learning in the field of sentiment analysis.
To better understand the emotions contained in multimodal data, these traits were easily combined with NLP
and contextual analysis. Proposed system is compared with SVM and random forest where VGG-16 CNN is
proved excellent i.e. accuracy 95 %, precision 95%, recall 93%, and F1 score 94%. Affective computing,
multimodal data integration, natural language creation, explainable artificial intelligence, and privacy-
preserving approaches are all on the horizon for sentiment analysis, which bodes well for its bright future.
These advancements have the potential to radically alter the field of mental healthcare by providing patients
with more tailored, compassionate care that also respects their right to privacy.

REFERENCES
[1] A. Yadav and D. K. Vishwakarma, “A comparative study on bio-inspired algorithms for sentiment analysis,” Cluster Computing,
vol. 23, no. 4, pp. 2969–2989, 2020, doi: 10.1007/s10586-020-03062-w.
[2] A. Deniz, M. Angin, and P. Angin, “Evolutionary multiobjective feature selection for sentiment analysis,” IEEE Access, vol. 9, pp.
142982–142996, 2021, doi: 10.1109/ACCESS.2021.3118961.
[3] J. Cui, Z. Wang, S. B. Ho, and E. Cambria, “Survey on sentiment analysis: evolution of research methods and topics,” Artificial
Intelligence Review, vol. 56, no. 8, pp. 8469–8510, 2023, doi: 10.1007/s10462-022-10386-z.
[4] S. M. A. -Ghuribi, S. A. M. Noah, and S. Tiun, “Unsupervised semantic approach of aspect-based sentiment analysis for large-scale
user reviews,” IEEE Access, vol. 8, pp. 218592–218613, 2020, doi: 10.1109/access.2020.3042312.
[5] N. Chen, Y. Sun, and Y. Yan, “Sentiment analysis and research based on two‐channel parallel hybrid neural network model with
attention mechanism,” IET Control Theory & Applications, vol. 17, no. 17, pp. 2259-2267, Apr. 2023, doi: 10.1049/cth2.12463.
[6] P. Nandwani and R. Verma, “A review on sentiment analysis and emotion detection from text,” Social network analysis and
mining, vol. 11, no. 81, 2021, doi: 10.1007/s13278-021-00776-6
[7] H. Liu, I. Chatterjee, M. Zhou, X. S. Lu, and A. Abusorrah, “Aspect-based sentiment analysis: a survey of deep learning methods,”
IEEE Transactions on Computational Social Systems, vol. 7, no. 6, pp. 1358–1375, Dec. 2020, doi: 10.1109/tcss.2020.3033302.
[8] L. Stappen, A. Baird, E. Cambria, B. W. Schuller, and E. Cambria, “Sentiment analysis and topic recognition in video
transcriptions,” IEEE Intelligent Systems, vol. 36, no. 2, pp. 88–95, 2021, doi: 10.1109/MIS.2021.3062200.
[9] M. S. Sulaiman, F. A. Maskan, Z. Derasit, and N. H. I. Teo, “Exploring YouTube comments to understand public sentiment on
covid-19 vaccines through deep learning-based sentiment analysis,” Malaysian Journal of Applied Sciences, vol. 8, no. 2, pp. 13–
27, Oct. 2023, doi: 10.37231/myjas.2023.8.2.353.
[10] S. Sachin, A. Tripathi, N. Mahajan, S. Aggarwal, and P. Nagrath, “Sentiment analysis using gated recurrent neural networks,” SN
Computer Science, vol. 1, no. 2, 2020, doi: 10.1007/s42979-020-0076-y.
[11] A. Yenkikar, C. N. Babu, and D. J. Hemanth, “Semantic relational machine learning model for sentiment analysis using cascade
feature selection and heterogeneous classifier ensemble,” PeerJ Computer Science, vol. 8, 2022, doi: 10.7717/PEERJ-CS.1100.
[12] Sugiyarto, J. Eliyanto, N. Irsalinda, and M. Fitrianawati, “Fuzzy sentiment analysis using convolutional neural network,” AIP
Conference Proceedings, vol. 2329, no. 1, Jan. 2021, doi: 10.1063/5.0042144.
[13] A. Yenkikar, C. N. Babu, and D. J. Hemanth, “SENTINET: a deep sentiment analysis network for political media bias detection,”
International Conference on Advancement in Computation & Computer Technologies (InCACCT), vol. 97, no. 6, pp. 645–651,
2022, doi: 10.6036/10593.
[14] Z. Jiang, W. Zaheer, A. Wali, and S. A. M. Gilani, “Visual sentiment analysis using data-augmented deep transfer learning
techniques, ” Multimedia Tools and Applications, vol. 83, pp. 17233–17249, 2024, doi: 10.1007/s11042-023-16262-4.
[15] I. E. Agbehadji and A. Ijabadeniyi, “Approach to sentiment analysis and business communication on social media,” in Bio-inspired
Algorithms for Data Streaming and Visualization, Big Data Management, and Fog Computing, pp. 169–193, Jan. 2021, doi:
10.1007/978-981-15-6695-0_9.
[16] M. Bansal, M. Kumar, M. Sachdeva, and A. Mittal, “Transfer learning for image classification using VGG19: Caltech-101 image
data set,” Journal of Ambient Intelligence and Humanized Computing, vol. 14, no. 4, pp. 3609–3620, 2023, doi: 10.1007/s12652-
021-03488-z.
[17] P. K. Chaubey et al., “Sentiment analysis of image with text caption using deep learning techniques,” Computational Intelligence
and Neuroscience, vol. 2022, pp. 1–11, Jun. 2022, doi: 10.1155/2022/3612433.
[18] S. Y. Chaganti, I. Nanda, K. R. Pandi, T. G. Prudhvith, and N. Kumar, “Image classification using SVM and CNN,” 2020
International Conference on Computer Science, Engineering and Applications, ICCSEA, pp. 1-5, 2020, doi:
10.1109/ICCSEA49143.2020.9132851.
[19] S. Sowden, B. A. Schuster, C. T. Keating, D. S. Fraser, J. L. Cook , “The role of movement kinematics in facial emotion expression
production and recognition,” Emotion, vol. 21, no. 5, Mar. 2021, doi: 10.1037/emo0000835.supp.
[20] S. Min, L. Yang, Z. Zhengfei, and Y. Dawei, “Sentiment analysis based on parallel hybrid network and attention mechanism,”
Journal of Computer Applications, vol. 40, no. 9, pp. 2543-2548, 2020, doi: 10.11772/j.issn.1001-9081.2019112020.
[21] P. Mehta, G. R. Chillarge, S. D. Sapkal, G. R. Shinde, and P. S. Kshirsagar, “Inclusion of children with special needs in the
educational system, artificial intelligence (AI),” AI-Assisted Special Education for Students With Exceptional Needs, pp. 156–185,

Elevating sentiment analysis with deep convolutional neural network model facial … (Pradnya Samit Mehta)
3402  ISSN: 2252-8938

2023, doi: 10.4018/979-8-3693-0378-8.ch007.

[22] K. Roman, “Facial emotion recognition dataset,” Kaggle, 2023. [Online]. Available:
https://www.kaggle.com/datasets/tapakah68/facial-emotion-recognition
[23] O. Khajuria, R. Kumar, and M. Gupta, “Facial emotion recognition using CNN and VGG-16,” 6th International Conference on
Inventive Computation Technologies, ICICT 2023 - Proceedings, pp. 472–477, 2023, doi: 10.1109/ICICT57646.2023.10133972.
[24] Y. Wang, Y. Li, Y. Song, and X. Rong, “The influence of the activation function in a convolution neural network model of facial
expression recognition,” Applied Sciences, vol. 10, no. 5, 2020, doi: 10.3390/app10051897.
[25] G. Chandrasekaran, N. Antoanela, G. Andrei, C. Monica, and J. Hemanth, “Visual sentiment analysis using deep learning models
with social media data,” Applied Sciences, vol. 12, no. 3, 2022, doi: 10.3390/app12031030.
[26] P. Mehta, S. Narwadkar, G. Chillarge, S. Rathi, G. Shinde, and C. Shewale, “A new approach to brain tumor detection with CNNS :
addressing the issues of standardization and generalizability,” International Journal of Intelligent Systems and Applications in
Engineering, vol. 12, no. 8, pp. 1–13, 2024.
[27] J. Singh and P. Tripathi, “A new approach to brain tumor detection with CNNS: addressing the issues of standardization and
generalizability,” 2021 10th IEEE International Conference on Communication Systems and Network Technologies (CSNT), 2021,
pp. 193-198, doi: 10.1109/CSNT51715.2021.9509679.

BIOGRAPHY OF AUTHORS

Pradnya Samit Mehta working at the Vishwakarma Institute of Information

Technology Pune. She received the B.E. and M.E. degree in Computer Engineering from SPPU
University, Pune, Maharashtra. She has 15 years teaching experience and 2 years of industrial
experience. Her research interests include artificial intelligence, deep learning, high performance
computing, and machine learning. She has published 20 papers in national and international
journals and conferences, 7 book chapters, 7 patents and 2 patents are granted. She can be
contacted at email: [email protected].

Renuka Wankhade currently an undergraduate student at Vishwakarma Institute of

Information Technology, is working towards a Bachelor's degree in Computer Science and
Engineering with a special focus on artificial intelligence and machine learning. She is a
passionate and dedicated learner. Moreover, she holds a deep fascination for deep learning and
generative AI, making these fields areas of immense personal interest. She can be contacted at
email: [email protected].

Dev Pankaj Chhabada is student at Vishwakarma Institute of Information

Technology, Pune Maharashtra. Pursuing B.Tech. in Artificial Intelligence and Data science.
Passionate about artificial intelligence, data science as well as machine learning. Also engaged
with a research project with HCLtech Pune Maharashtra. Keen Interest includes learning
techniques of machine learning and applying them to solve real life problems and applications.
He can be contacted at email: [email protected].

Dhimahi Patel is a dedicated student currently pursuing a Bachelor of Technology

(B.Tech.) degree in Artificial Intelligence and Data Science at Vishwakarma Institute of
Information Technology, Pune. Her academic journey is characterized by a fervent passion
for artificial intelligence, data science, and machine learning. Her enthusiasm for exploring
cutting-edge technologies and their practical applications is a driving force in her pursuit of
solutions to real-world problems, and she continually seeks opportunities to contribute to the ever-
evolving landscape of AI research and application. She can be contacted at email:
[email protected].

Int J Artif Intell, Vol. 13, No. 3, September 2024: 3395-3403

Int J Artif Intell ISSN: 2252-8938  3403

Anirudh Kailas Gote a dedicated student of Artificial intelligence and data science
at Vishwakarma Institute of Information Technology (VIIT), demonstrates an aptitude for
visionary thinking and a natural ability to foster connections within his community. His academic
journey is characterized by a fervent passion for artificial intelligence, data science, and machine
learning, reflecting his enthusiasm for exploring cutting-edge technologies and their practical
applications. Passionate about research, he is currently engaged in a project with HCl technologies
and previously completed an internship at Beepsy. He can be contacted at email:
[email protected].

Anuradha Yenkikar received Master’s degree in IT from Savitribai Phule Pune

University, Pune and completed Ph.D. from Ramaiah University of Applied Sciences, Bangalore.
She has 11 years of teaching experience. She has published 20 papers in various international
journals and conferences. She has a filed a patent to her name. Her current research interests
include machine learning, deep learning, and parallel computing. She has completed a research
project funded by Savitribai Phule Pune University. She can be contacted at email:
[email protected].

Poorva Agrawal obtained her P.hD. degree in Computer Science and Engineering
from Symbiosis International (Deemed University), Pune, India in 2020. She obtained her
M.Tech. degree in Computer Science and Engineering from SGSITS, Indore, India in 2012. She
has been engaged in research and teaching for more than 12 years. At present she is working as
Senior Assistant Professor in Department of Computer Science and Engineering at Symbiosis
Institute of Technology Nagpur, Symbiosis International (Deemed University) Pune, MH, India.
She has presented more than 20 papers in international/national journals/conferences. Her
research interests include artificial intelligence, machine learning, data science, image processing,
and computer vision. She can be contacted at email: [email protected].

Gagandeep Kaur completed her Ph.D. degree in Computer Science & Engineering
from LPU, Punjab, India. She is currently working as Assistant Professor in Department of
Computer Science at Symbiosis Institute of Technology, Nagpur. She has more than 11 years of
experience in teaching. Her area of interests includes NLP, AI/ML, data science, and image
processing. She can be contacted at email: [email protected].

Elevating sentiment analysis with deep convolutional neural network model facial … (Pradnya Samit Mehta)