=Paper=
{{Paper
|id=Vol-3682/Paper17
|storemode=property
|title=AWS-Enhanced Sentiment Analysis Using LSTM For Online
Video Comments
|pdfUrl=https://ceur-ws.org/Vol-3682/Paper17.pdf
|volume=Vol-3682
|authors=P Nandieswar Reddy,Sai Aswath S,Rithvika Alapati,Beena B M
|dblpUrl=https://dblp.org/rec/conf/sci2/ReddySAM24
}}
==AWS-Enhanced Sentiment Analysis Using LSTM For Online
Video Comments==
https://ceur-ws.org/Vol-3682/Paper17.pdf
AWS-Enhanced Sentiment Analysis Using LSTM For Online
Video Comments⋆
P Nandieswar Reddy1,∗,†, Sai Aswath S1,†,Rithvika Alapati1,† and Dr. Beena B.M.1,†
1Department of Computer Science & Engineering, Amrita School of Computing, Amrita Vishwa Vidyapeetham,
Bangalore, Karnataka, 560035, India
Abstract
Sentimental analysis is critical in understanding the user’s reaction toward the content on social
media platforms. YouTube is one of the most used social media platforms in the current era.
Understanding the user’s reaction towards the content posted on such platforms is important in
improving the content. A sentimental model using LSTM and NLP techniques is built and trained
using the IMDB dataset and deployed using Amazon web services (AWS). 85% accuracy persisted
and detailed the model’s performance in categorizing comments as positive and negative. The
interactive dashboard is built using stream-lit.
Keywords
Amazon Web Services (AWS), Sentimental Analysis, Cloud Computing, LSTM, NLP, Deep
Learning.1
1. Introduction
In today’s virtual surroundings, user-generated content material on systems together with
YouTube has turned out to be a quintessential part of online communication. The number
of comments and the kind of remarks consisting of video reflect customers’ rich feelings.
Emotion evaluation, a developing practice in natural language processing, affords a method
of decoding the underlying emotional tones that underlie those troubles. This application is
pushed using the want to apply sentiment evaluation to YouTube content, unpacking the
emotions expressed by customers and supplying actionable insights to content material
creators and platform managers. The main motto for stepping into YouTube comment
sentiment analysis lies in its capability to transform content strategy, community
engagement, and platform dynamics, and a video posted on YouTube will have millions of
comments, and it is very hard for the creator to go through all of them and understand the
user requirements.
Symposium on Computing & Intelligent Systems (SCI), May 10, 2024, New Delhi, INDIA
∗ Corresponding author.
† These authors contributed equally.
nandieswar.pelleti123@gmail.com (P.N. Reddy); saiaswath48@gmail.com (S. Aswath);
rithvikaalapati1@gmail.com (R. Alapati) ; bm_beena@blr.amrita.edu(Dr. B. B.M.)
0009-0008-8499-5602 (P. N. Reddy); 0009-0001-5280-9982 (S. Aswath); 0009-0003-7390-3972 (R.
Alapati) ; 0000-0001-9108-7073 (Dr. B. B.M.)
© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
Exploring audience emotion in their content can help creators customize their content
to resonate more with the target content, leading to elevated engagement and viewership.
Also, it enables corporations to read purchaser sentiment toward their products or services
and may offer precious comments for improving product development, advertising
techniques, and customer service. Research scholars can examine public sentiment on
numerous subjects and may contribute to investigations in psychology, politics, advertising,
and marketing. As motion pictures acquire thoughts and memories, information, and the
emotions expressed in that feedback are essential for content creators trying to align their
content material with target audience options, and platform managers trying to own
experience may be superior. As a part of cloud computing, Amazon Web Services (AWS)
offers multiple responses that combine nicely with our sentiment assessment framework in
YouTube contexts. It translates into advanced overall performance, scalability, and global
get-right of access.
The versatility of AWS is clear in its scalability competencies, which is a high problem
given the dynamic nature of YouTube content. With the capability to dynamically scale assets
primarily based on calls, AWS ensures that our sentiment assessment model remains responsive,
even at some point of durations of increased times whilst clients are connected. Efficiency in
model learning is a cornerstone to carrying out our task, and AWS allows it fairly in this aspect.
With parallel processing abilities, AWS hurries up the training of our sentiment assessment
model, which is a specially complicated venture dealing with large datasets that include IMDb.
Now, this parallelization no longer reduces schooling issues but supports the iterative
refinement technique, if we suppose it is suitable and tremendous. Additionally, AWS’ managed
offerings, mixed with Sage Maker, play a key feature in streamlining our tool studying workflow.
By abstracting the complexity of infrastructure management, AWS shall see the evolution and
optimization of our sentiment analysis version. Scalability and cost-effectiveness enable us to
optimize using AWS content material and adapt sensitivity evaluation obligations to the right
computational goals. Given YouTube’s global target market, global reach is paramount. AWS’s
global community of directory services, blended with Content Delivery Network services,
ensures the reach of our sentiment analysis software program. This global strategy contributes
to continuing consumer liberty, irrespective of the geographical place of users interacting with
our application. Additionally, AWS offers a reliable and robust environment for our sentiment
analysis software program. The chosen approach involves training a sentiment analysis version
using the IMDb dataset, which is a comprehensive film analysis repository. Using the skills of
Amazon Web Services (AWS), the model is skilled in using sentiment analysis concepts from
movie reviews to diverse kinds of content determined in YouTube content material. This fact’s
structure and platform choice ensures a foundation of tough for the version, growing its
flexibility and performance.
In the upcoming sections, we delve into a comprehensive literature review, explore AWS
services, outline our methodology, detail the implementation process, present our findings, and
conclude with insights drawn from the analysis. Through this endeavor, we aim to contribute to
the advancement of sentiment analysis in the context of user-generated content on platforms
like YouTube, thereby enhancing content creation, audience engagement, and platform
dynamics.
2. Related Work
Several studies have explored sentiment analysis on YouTube comments, shedding light
on the diverse approaches and insights gained from analyzing viewer sentiments. One
paper proposes a method to forecast the like ratio of YouTube videos by analyzing the
emotive tone of viewer comments using sentiment analysis. The method involves
preprocessing the comments, categorizing them as positive, negative, or neutral, and
estimating the like ratio based on the percentage of positive comments [1]. Another study
focuses on sentiment analysis of YouTube comments related to the construction of the
Mengwi-Gilimanuk Toll Road in Bali Province during the Jokowi era. Utilizing the naïve
Bayes algorithm, the research evaluates opinions and reveals varying emotional
distributions within comments, providing valuable insights into public sentiment [2]. An
analysis of YouTube comments on the Kompas TV channel investigates public sentiment
towards potential candidates for the 2024 Indonesia presidential election. Using Python
libraries such as Pandas, matplotlib, wordcloud, and textblob, the study reveals positive
sentiments towards specific candidates, offering insights into voter preferences [3]. A
research paper introduces an NLP-based model to classify Arabic comments on YouTube as
positive or negative, achieving high accuracy with the Naïve Bayes classifier. This study
bridges a literature gap in Arabic sentiment analysis, providing valuable insights for content
creators aiming to improve audience engagement [4]. Research data from a YouTube
channel for the 2019 presidential election debate comprises 31,947 comments, balanced
using oversampling. Skip-gram is utilized for feature extraction, and Random Forest is
employed for sentiment classification. The study sheds light on the sentiment distribution
among viewers regarding political debates [5]. The focus of another paper is sentiment
analysis using Amazon Web Services (AWS) on Twitter data, managing data on AWS Elastic
Compute Cloud (EC2) with elastic load balancing. The proposed logistic regression model
achieves high accuracy, surpassing existing algorithms, and highlights the effectiveness of
advanced machine learning techniques in sentiment analysis on AWS [6]. The importance
of load balancing in cloud computing is emphasized in a study, outlining types and
techniques for distributing workload among nodes effectively, contributing to the
optimization of resource usage in cloud environments [7].
A paper introduces a Levenshtein distance-based sentiment classification engine
analyzing product reviews to aid users in making informed choices, showcasing the
application of advanced techniques in sentiment analysis [8]. Sentiment analysis of Twitter
data is explored in another paper, comparing the performance of machine learning
algorithms on different datasets, providing insights into the effectiveness of various
approaches in sentiment analysis [9]. A sentiment analysis-based video classification
system is proposed, categorizing YouTube videos into abusive and non-abusive categories
using techniques such as Bag of Words and logistic regression, offering a solution for
identifying and managing abusive content on online platforms [10]. Methods and
techniques for sentiment analysis of YouTube comments are discussed in a study,
emphasizing their relevance in data mining and sentiment analysis research, and providing
insights into the challenges and opportunities in analyzing user-generated content [11].
Automated sentiment analysis of real-time YouTube comments on the TV show "Game of
Thrones" is proposed in one paper, showcasing the application of sentiment analysis in
understanding user reactions to popular media content [12]. Another paper focuses on
sentiment analysis of YouTube video comments, achieving high accuracy levels with Naïve
Bayes and Support Vector Machine classifiers, highlighting the effectiveness of machine
learning algorithms in sentiment analysis tasks [13]. These studies collectively highlight the
importance of sentiment analysis in understanding viewer engagement and provide
valuable insights for content creators and platform managers.
3. AWS Services
In this study, we present the utilization of various Amazon Web Services (AWS) offerings
to develop a sentiment analysis system. Leveraging the capabilities of AWS, we demonstrate
the streamlined implementation of machine learning models for sentiment analysis tasks.
The study focuses on integrating AWS services to preprocess data, train machine learning
models, deploy endpoints, and create a user-friendly web interface for interaction.
Amazon Sage Maker, a fully managed service, played a pivotal role in training the
sentiment analysis model using Long Short-Term Memory (LSTM). By offering a managed
environment for model development, training, and deployment, Sage Maker streamlined
the machine learning workflow, facilitating efficient model iteration and experimentation.
Amazon S3 served as the primary data storage solution in the project. It was utilized for
storing both the preprocessed data and the trained sentiment analysis model. Leveraging
its scalability, data availability, and security features, S3 provided a centralized and
accessible location for data storage, enabling seamless data management and analysis.
Amazon API Gateway facilitated the creation of a RESTful API that served as a
communication bridge between the frontend and backend components of the sentiment
analysis system. By creating, publishing, and managing APIs at scale, API Gateway ensured
seamless interaction between the Streamlit web app, and the backend Lambda functions
responsible for making predictions.
AWS Lambda, a serverless computing service, played a crucial role in executing
predictions using the trained sentiment analysis model. By running code without
provisioning or managing servers, Lambda functions dynamically scaled based on incoming
workload, ensuring efficient resource utilization and cost-effectiveness in processing user
requests from the Streamlit web app.
Amazon EC2 instances were utilized to orchestrate the sentiment analysis workflow,
managing tasks such as data preprocessing, model training, and storage processes. Offering
secure, resizable compute capacity in the cloud, EC2 instances ensured a persistent and
scalable computing environment, enhancing the overall reliability and performance of the
sentiment analysis system.
In addition to leveraging AWS services, a Streamlit web application was developed as the
front-end interface for the sentiment analysis system. This interactive and intuitive web app
allowed users to input either text or video data for sentiment analysis. The app seamlessly
communicated with the backend components, utilizing the RESTful API created with
Amazon API Gateway to facilitate sentiment analysis predictions. By providing a dynamic
and user-friendly experience, the Streamlit web app enhanced the accessibility and usability
of the sentiment analysis system, empowering users to interact with the underlying
functionalities effectively.
4. Methodology
The methodology in the development of the sentiment analysis or reaction analysis
model for YouTube comments is designed to extract meaningful insights from user
comments in any YouTube video. The process begins with data assembling, where reviews
of content are labeled. This initial step emphasizes the importance of ethical data collection
rules. Once the dataset is assembled, the next step involves data preprocessing. This step is
crucial to refining the collected data, involving removing extra information such as stop
words, punctuation, and special characters. Converting the text data to a suitable format
deep-learning models is also undertaken to prepare the dataset for effective model training.
An important part of the methodology lies in training the sentiment analysis model. The
choice of a specific model is important for the characteristics of the dataset and for obtaining
better accuracy. The dataset is split into training and validation sets for model training to
ensure the model's performance across different data scenarios.
Figure 1: High-level architecture
After model training, a crucial step involves testing the model on a different dataset.
Evaluation metrics such as accuracy, precision, recall, and F1 score are measured to
evaluate the model's ability to new, unseen data. Upon successful training of the sentiment
analysis model, the subsequent step involves its deployment, a pivotal phase in making the
model available for real-world application. In the deployment process, the model is
seamlessly integrated into a YouTube comment analysis application or website, allowing
users to access sentiment insights in real-time. Notably, the deployment leverages cloud
computing infrastructure, specifically Amazon Web Services (AWS), for several compelling
reasons. Deploying the sentiment analysis model on the cloud, and more specifically on
AWS, matches with the goal of the project's commitment to scalability, efficiency, and global
accessibility. The scalability of AWS makes sure that the model can handle varying loads,
adapting to the dynamic nature of YouTube comments where levels of engagement will
rapidly fluctuate. The cloud environment provides the efficient allocation of resources
based on demand, optimizing performance for the greater number of YouTube comments.
Moreover, the use of AWS for deployment offers cost-effective solutions. The pay-as-you-go
pricing model makes sure that the project costs are proportional to the actual resources
utilized during model deployment and inference. This cost efficiency aligns seamlessly with
the project's budget constraints and underscores the advantage of cloud-based deployment
for resource optimization. AWS's global infrastructure, including a network of data centers
worldwide, contributes to the low-latency access of the sentiment analysis application. This
global accessibility is paramount, considering the diverse and international user base of
YouTube. Furthermore, the Content Delivery Network services provided by AWS enhance
the rapid delivery of sentiment analysis results, ensuring a seamless and responsive user
experience across different geographical locations. Security considerations are paramount
during deployment, and AWS provides robust security measures. Encryption protocols and
access control mechanisms are implemented to safeguard both the deployed sentiment
analysis model and the user data processed by the application. The trusted security features
of AWS ensure the confidentiality and integrity of the deployed system. Incorporating
cloud-based deployment through AWS not only enhances the scalability, efficiency, and
global accessibility of the sentiment analysis model but also aligns with contemporary best
practices in machine learning deployment. Cloud integration ensures that the sentiment
analysis application remains adaptable to the evolving landscape of YouTube comments
while providing a reliable, cost-effective, and secure solution for real-time analysis. The
high-level architecture of the methodology is shown in Figure 1.
5. Implementation
In a sentimental analysis system, the implementation journey begins with data
collection. The IMDB dataset, also known as the Large Movie Review Dataset v1.0, serves as
an extensive resource specifically designed for binary sentiment classification. It
encompasses a total of 50,000 movie reviews, meticulously categorized into equal halves of
25,000 positive and 25,000 negative reviews. These reviews are composed in English and
stored as individual text files, exhibiting a diverse range of sizes ranging from 1 kilobyte to
15 kilobytes. This variability in file sizes provides a rich set of textual lengths, facilitating
thorough analysis. Importantly, the text files intentionally omit any rating information,
focusing solely on the narrative content of the reviews. The complete dataset description is
mentioned in Table 1.
Table 1
Dataset Description.
Attribute Description
Name IMDB Dataset
Positive 25000
reviews
Negative 25000
reviews
Language English
File Format Text Files
Training Set 25000
Testing Set 25000
IMDb ratings typically span from 1 to 10, and the dataset creator has established specific
criteria for sentiment labeling. Reviews with ratings of 4 stars or lower are categorized as
negative, while those with ratings of 7 stars or higher are identified as positive. Reviews
falling outside these rating ranges are deliberately excluded from the dataset. The training
set comprises the raw text of 25,000 IMDb movie reviews, each explicitly marked as either
positive or negative. This intentional balance ensures a fair distribution for training
machine learning models in the domain of sentiment analysis. In contrast, the test set
consists of 25,000 unlabeled movie reviews, presenting a challenge for sentiment prediction
without explicit class labels.
This unlabeled set serves as a valuable tool for researchers and practitioners, allowing
them to assess the generalization capabilities of models to previously unseen data. In the
data preprocessing and loading pipeline for the IMDB dataset, specifically designed for
sentiment analysis tasks, each review in the training and test sets is labeled as either
positive or negative based on the IMDb rating system. This labeling ensures that the
sentiment of each review is explicitly denoted, facilitating supervised learning for sentiment
analysis models. To gain insights into the dataset's distribution, an analysis is performed to
understand the balance between positive and negative reviews in both the training and test
data. This step is crucial for assessing the dataset's representativeness and its potential
impact on model training and evaluation. The data is then shuffled to create balanced and
randomized training and test sets. This randomization helps prevent any bias that may arise
from the original ordering of reviews, ensuring a more robust training and evaluation
process for machine learning models. As part of the preprocessing steps, HTML tags are
removed from the text using the Beautiful Soup library. The text is converted to lowercase
to ensure uniformity, tokenized for further analysis, and common English stop words (e.g.,
"the", “and” "is") are eliminated. Removing stop words is beneficial as they often do not
contribute significantly to the overall meaning of the text. Additionally, stemming is applied
using the Porter Stemmer to reduce words to their root form. This process helps in
consolidating similar words, contributing to the efficiency of the subsequent analysis.
Furthermore, any characters that are not alphanumeric are removed from the pre-
processed data. This step ensures that the data is clean and focuses solely on meaningful
content, enhancing the quality of the analysis. Finally, the pre-processed data is uploaded to
an S3 bucket, providing a centralized and accessible location for further analysis and model
training. This well-defined and thorough preprocessing pipeline sets the stage for effective
sentiment analysis. The next crucial step involves vectorization using word frequency. This
process transforms the textual data into numerical vectors, representing the frequency of
each word. The resulting vectorized dataset is then arranged in descending order, capturing
the importance of words based on their occurrence frequency. This structured dataset is
instrumental in training our sentiment analysis model, providing a foundation for
understanding the underlying sentiments within YouTube comments. In sentiment
analysis, the model training phase is a critical step, and we leverage the capabilities of
Amazon Sage Maker to streamline this process. The selected model architecture is LSTM, a
recurrent neural network (RNN) known for its proficiency in capturing sequential
dependencies within textual data.
The embedding dimension is set at 32, representing the size of the vector space in which
words are embedded. The hidden dimension, set to 100, determines the size of the LSTM's
hidden state, influencing its capacity to capture and retain information from input
sequences. The vocabulary size is capped at 5000, defining the number of unique words
considered during training. This limitation manages computational complexity while still
accommodating a diverse range of words. The loss function employed is Binary Cross-
Entropy (BCE) Loss, a suitable choice for binary sentiment classification tasks. The
optimizer chosen is Adam, known for its adaptive learning rates and efficient convergence
during optimization. The learning rate is maintained at its default value to strike a balance
between model convergence and computational efficiency. The training process spans 20
epochs, ensuring an adequate number of passes through the dataset for effective learning
without risking overfitting. A batch size of 50 is utilized during training, influencing the
number of samples processed in each iteration. The architecture and parameters of the
model are mentioned in Table 2.
Table 2
Model Architecture and Parameters.
Parameter Value
Embedding 32
dimensions
Hidden 100
dimensions
Dense layer 1
Loss Function BCE loss
Optimizer Adam
Learning rate 0.001
Epochs 20
Batch size 50
Utilizing Amazon Sage Maker offers several advantages in this context. Firstly, Sage
Maker simplifies the entire machine-learning workflow, providing a managed environment
for model development, training, and deployment. It allows for seamless integration with
other AWS services, facilitating data storage, preprocessing, and deployment. The
scalability of Sage Maker accommodates varying workloads, ensuring efficient resource
utilization during the training phase. Additionally, Sage Maker provides a secure and
controlled environment for model development, addressing concerns related to access
control and data security. The trained model is stored as an endpoint. By defining IAM roles
with specific permissions, we establish a secure environment that governs who or what can
access the endpoint storage. Identity and Access Management (IAM) plays a crucial role in
ensuring secure and controlled access to AWS resources. IAM is particularly vital when
dealing with the process of saving the trained model's endpoint. This access control
mechanism prevents unauthorized modifications or access to critical components of the
model, safeguarding the integrity and security of the sentiment analysis system. IAM roles
are configured to grant the necessary permissions for saving the trained model's endpoint
securely. When a prediction request is made, the Lambda function is triggered, invoking the
sentiment analysis model stored on the Sage Maker endpoint. AWS Lambda functions serve
as the backbone for executing predictions using the saved Sage Maker endpoint in our
sentiment analysis system. Lambda functions, being serverless, offer a scalable and cost-
effective solution for on-demand computation. This architecture ensures efficient resource
utilization, as the Lambda functions dynamically scale based on the incoming workload.
Amazon API Gateway serves as a central communication hub, creating a RESTful API that
connects the frontend and backend components of our sentiment analysis system. This API
facilitates seamless interaction, allowing the Streamlit web app to communicate with the
Lambda functions responsible for making predictions. The API Gateway also plays a crucial
role in ensuring that the various components of our system can efficiently exchange data,
contributing to a cohesive and well-orchestrated system architecture. Using RESTful APIs
provided by Amazon API Gateway ensures standardized communication protocols and
enables easy integration between different components. This not only simplifies the
development process but also enhances the maintainability and scalability of our sentiment
analysis system. The API Gateway acts as a bridge, ensuring smooth data flow and effective
communication between the front end and back end, ultimately contributing to a user-
friendly and efficient application. Amazon Elastic Compute Cloud (EC2) instances take on
the role of orchestrating the entire sentiment analysis workflow.
Figure 2: AWS Architecture
These instances manage critical tasks such as data preprocessing, model training, and
storage processes, providing a centralized environment for streamlined execution. The
orchestration capabilities of EC2 ensure that each component of the system functions
cohesively, contributing to the overall efficiency of the sentiment analysis pipeline. EC2
instances are particularly advantageous for tasks that demand a persistent and scalable
computing environment. In our case, EC2 plays a key role in managing the workflow,
ensuring that the various stages of sentiment analysis are executed in a coordinated
manner. This orchestration enhances the overall reliability and performance of our system,
aligning with best practices in machine learning workflows. The front end of our sentiment
analysis system is developed using Streamlit, offering an intuitive and interactive user
interface. Users can input either text or video, and the frontend seamlessly communicates
with the backend components to facilitate sentiment analysis. The text input allows users
to input statements, receiving prompt sentiment outputs, while video input enables users
to input YouTube video links for comprehensive sentiment analysis of the associated
comments. The Streamlit web app provides a dynamic and user-friendly experience, making
it easy for users to interact with the sentiment analysis system. The front end not only
ensures a smooth user experience but also serves as a crucial component in connecting
users to the underlying sentiment analysis functionalities. By providing a clear and intuitive
interface, the front end enhances the accessibility and usability of our sentiment analysis
system.
6. Results
The evaluation metrics for our sentiment analysis application deployed in the AWS
platform demonstrate its effectiveness in correctly classifying both positive and negative
comments. Precision, representing the percentage of comments predicted to be positive
that are positive, attains commendable values of 0.92 for positive comments and 0.80 for
negative comments. These scores indicate that the model is adept at accurately identifying
both positive and negative sentiments within the comments. The recall metric, indicating
the percentage of actual positive comments that the model correctly classified as positive,
presents values of 0.77 for positive comments and an impressive 0.93 for negative
comments. While the model is slightly less likely to correctly identify positive comments, it
excels in identifying negative sentiments, showcasing a robust capability to capture various
nuances in sentiment expressions. The F1-score, a harmonized average of precision and
recall, provides a comprehensive assessment of the model's ability to correctly identify both
positive and negative comments. For positive comments, the F1- score is 0.84, and for
negative comments, it is 0.86.
Figure 3: Text input in the frontend Figure 4: Video input in the frontend
These scores reflect a balanced performance across precision and recall, suggesting that
the model maintains a good equilibrium in correctly classifying sentiments in the
comments. With an overall accuracy of 0.85, these evaluation metrics collectively affirm the
strong performance of our sentiment analysis model. The high precision, recall, and F1-
score values underscore its proficiency in effectively distinguishing between positive and
negative sentiments within the YouTube comments, contributing to an accurate and reliable
sentiment analysis system. The main implementation results are seen in the front-end part
which has two types of input, that is text as displayed in Figure 3. The video input is shown
in Figure 4. The video input should be given as a YouTube link and the number of comments
should be selected then the results will be displayed as shown in Figure 5. Finally, the
statistics of the count of positive and negative comments will be displayed in the pie chart
as shown in Figure 6.
Figure 5: Results of the video input Figure 6: Pos/Neg Comments Pie Chart
Developing a sentiment analysis model using LSTM and NLP techniques on the IMDB
dataset and deploying it on AWS with an 85% accuracy rate presents a valuable tool for
enhancing content quality and user experience on platforms like YouTube. By
understanding user sentiments, content creators can tailor their videos to better meet
audience preferences, leading to more engaging and relevant content. However, limitations
such as the model's inability to accurately interpret sarcasm, slang, or context-dependent
language nuances may affect its effectiveness in certain scenarios. Additionally, biases in the
training data or model architecture could result in skewed sentiment analysis results.
Despite these limitations, the tool provides an efficient feedback loop for content creators,
fosters community engagement, and aids in the prevention of harmful content
dissemination. Integrating an interactive dashboard using Streamlit enhances accessibility
but may require ongoing maintenance and updates to ensure optimal performance and
usability.
7. Conclusion
YouTube comment sentiment analysis yielded highly encouraging results. The model
demonstrated remarkable accuracy, exceeding 85% in its ability to correctly classify both
positive and negative comments. This impressive performance is further underscored by strong
precision values: 0.92 for positive comments and 0.80 for negative comments. The model has an
exceptional performance in identifying negative sentiments, achieving a recall score of 0.93. It
also performs well in identifying positive comments, with a recall score of 0.77. This highlights
the model's ability to capture nuanced and subtle expressions within the comments.
To comprehensively evaluate the model's performance, we employed the F1-score metric,
which harmoniously balances precision and recall. The F1-scores of 0.84 for positive comments
and 0.86 for negative comments further solidify the model's balanced and effective classification
capabilities. Beyond mere metrics, the project boasts a user-friendly interface designed to
empower content creators. The interface accepts both video and text input options, offering
flexibility and convenience to users. The generated insights are presented clearly and concisely,
utilizing sentiment pie charts. This project's significance lies in its potential to revolutionize
content creation on YouTube. By equipping creators with the ability to accurately understand
audience sentiment, the model enables them to: Cultivate stronger audience relationships: By
actively engaging with viewers based on their expressed sentiments, creators can foster a more
positive and interactive community. Make data-driven content decisions: Insights gleaned from
sentiment analysis inform content creation strategies, ensuring that content aligns with
audience preferences and maximizes engagement. Gain a competitive edge: Understanding
audience sentiment empowers creators to stay ahead of the curve, tailoring their content to
resonate with their viewers and differentiate themselves from the competition.
In the future, the scope of sentiment analysis models using LSTM and NLP techniques
deployed on platforms like AWS extends to broader applications across various industries. These
models can be adapted to analyze sentiments not only in text but also in other forms of media
such as audio and video content. Additionally, advancements in deep learning and natural
language processing can lead to even more accurate and nuanced sentiment analysis, including
the identification of sarcasm, irony, and cultural nuances.
Furthermore, integrating sentiment analysis with recommendation systems can personalize
user experiences further, offering content suggestions based on sentiment preferences. As social
media platforms continue to evolve and diversify, sentiment analysis models will play a crucial
role in understanding user behavior, informing marketing strategies, and shaping online
interactions. Collaboration with interdisciplinary fields such as psychology and sociology can
also deepen our understanding of human emotions and behaviors in digital environments.
Ultimately, the future scope of sentiment analysis holds immense potential for enhancing user
experiences, promoting meaningful interactions, and contributing to a more informed and
inclusive digital society.
In the future, the scope of sentiment analysis models using LSTM and NLP techniques
deployed on platforms like AWS extends to broader applications across various industries. These
models can be adapted to analyze sentiments not only in text but also in other forms of media
such as audio and video content. Additionally, advancements in deep learning and natural
language processing can lead to even more accurate and nuanced sentiment analysis, including
the identification of sarcasm, irony, and cultural nuances.
Furthermore, integrating sentiment analysis with recommendation systems can personalize
user experiences further, offering content suggestions based on sentiment preferences. As social
media platforms continue to evolve and diversify, sentiment analysis models will play a crucial
role in understanding user behavior, informing marketing strategies, and shaping online
interactions. Collaboration with interdisciplinary fields such as psychology and sociology can
also deepen our understanding of human emotions and behaviors in digital environments.
Ultimately, the future scope of sentiment analysis holds immense potential for enhancing user
experiences, promoting meaningful interactions, and contributing to a more informed and
inclusive digital society.
References
[1] Sentimental Analysis of YouTube Video Comments Using Bagging Ensemble Learning
Approach. (2023).;02(04) doi: 10.55041/isjem00336
[2] I, Putu, Agus, Eka, Darma, Udayana., I., G., A., Indrawan., I, Putu, Denny, Indra, Putra.
Decision Support System for Sentiment Analysis of YouTube Comments on
Government Policies. Journal of Computer Networks, Architecture and High
Performance Computing, (2023).;5(1):27-37. doi: 10.47709/cnahpc.v5i1.1999
[3] Dimaz, Cahya, Ardhi., Dwi, Puspita, Sari. Sentiment Analysis of YouTube Comments:
Potential Indonesian Presidential Election Candidates. International Journal of
Computer Applications Technology and Research, (2022). 451-456.doi:
10.7753/ijcatr1112.1010
[4] Dhiaa, Musleh., Nasro, Min-Allah., Mamoun, Masoud, Abdulqader. Arabic Sentiment
Analysis of YouTube Comments: NLP-Based Machine Learning Approaches for Content
Evaluation. Big data and cognitive computing, (2023).;7(3):127-127. doi:
10.3390/bdcc7030127
[5] Siti, Khomsah. Sentiment Analysis On YouTube Comments Using Word2Vec and
Random Forest. Telematika, (2021).;18(1):61-72. doi:
10.31315/TELEMATIKA.V18I1.4493
[6] Theodoros, G., Iliopoulos. (2023). A Cloud-Based Sentiment Analysis through Logistic
Regression in AWS Platform. Computer Systems: Science & Engineering, 45(1):857-
868. doi: 10.32604/csse.2023.031321
[7] J. Uma, V. Ramasamy, A. Kaleeswaran, “Load Balancing Algorithms in Cloud computing
Environment - A Methodical Comparison”, in International Journal of Advanced
Research in Computer Engineering & Technology (IJARCET 2014) Volume 3, Issue
2, February 2014.
[8] Anjali T., T R Krishnaprasad, and P Jayakumar, “A Novel Sentiment Classification of
Product Reviews using Levenshtein Distance”, in 2020 International Conference on
Communication and Signal Processing (ICCSP), Chennai, India, 2020.
[9] S. Srihitha Yadlapalli, R. Reddy, R., and Sasikala T, “Advanced Twitter Sentiment
Analysis Using Supervised Techniques and Minimalistic Features”, in Ambient
Communications and Computer Systems, Singapore, 2020
[10] Debabrata, Swain., Monika, Verma., Sayali, Phadke., Shraddha, Mantri., Anirudha,
Kulkarni. "Video Categorization Based on Sentiment Analysis of YouTube Comments."
undefined (2021). doi: 10.1007/978-981-33- 4859-2_6
[11] Rawan, Fahad, Alhujaili., Wael, M.S., Yafooz. "Sentiment Analysis for Youtube Videos
with User Comments: Review." (2021). doi: 10.1109/ICAIS50930.2021.9396049
[12] Shivam, Sharma., Hemant, Kumar, Soni. "Viewer’s Sentiments on Game of Thrones: An
Automated Lexicon- Based Sentiment Analysis on Real-Time YouTube Comments."
2021. doi: 10.1007/978-981-16-1295-4_31
[13] Fiktor, Imanuel, Tanesab. "Sentiment Analysis Model Based on Youtube Comment
Using Support Vector Machine." (2017).