Enhancing YouTube Comment Insights: A Machine Learning

Approach to Sentiment Analysis

Dipika Raigar

, Poonam Chapke

, Pranav Dangare

, Pratik Dhagude

and Varsha Malode

Department of Computer Engineering, SCTR’s Pune Institute of Computer Technology, Pune-411043, Maharashtra, India

Keywords: YouTube Comments, Machine Learning, Natural Language Processing, Supervised Learning, Sentence

Classification.

Abstract: As YouTube channels grow, they receive large volumes of comments that provide valuable feedback, crucial

for understanding audience sentiment and improving engagement. However, existing sentiment analysis

approaches focus only on single-language positive/negative classification, and struggle with the informal

language in YouTube comments. This project addresses these limitations by building a system to classify

comments based on sentiment (positive, negative, neutral) and sentence types, using advanced NLP

techniques and machine learning models. Our system will support multilingual analysis, increasing

accessibility, and its performance will be evaluated using cross-validation and F1 scores to help creators

improve their contents.

1 INTRODUCTION

The rise of social media platforms like Facebook,

Twitter, and YouTube has changed how people

interact with content. These platforms allow users to

communicate, share ideas, and express opinions

openly. Among them, YouTube stands out as one of

the most popular video sharing platforms. It not only

gives users a chance to watch videos but also to share

their thoughts through comments. These comments

serve as feedback for the video creators, offering

them a way to understand how their audience feels

about their content. YouTube comments are essential

because they often reflect the public’s overall opinion

or mood. This feedback can help content creators

know what their audience likes or dislikes, allowing

them to improve their future videos and create content

that resonates better with viewers. However, with the

massive number of comments posted daily, it

becomes difficult to analyse and understand all of

them manually. This is where technology,

specifically Sentiment Analysis (SA), becomes very

https://orcid.org/0000-0002-9906-9459

https://orcid.org/0009-0009-3682-0952

https://orcid.org/0009-0009-1325-3115

https://orcid.org/0009-0007-2724-3829

https://orcid.org/0009-0000-1744-6410

useful. Sentiment Analysis is a branch of Natural

Language Processing (NLP) that focuses on

identifying the emotions or opinions expressed in a

piece of text. For instance, it can determine whether

a comment is positive, negative, or neutral.

However, analysing YouTube comments comes

with its own set of challenges. Unlike formal

writing, YouTube comments are usually

unstructured. People often use slang, abbreviations,

emojis, or even misspelled words, making it harder

to understand the real meaning behind the text.

Also, sentence structures can vary greatly from one

comment to another. Some comments might be long

and detailed, while others are short and vague.

These factors make it tough to process the

comments automatically using traditional tools. The

goal of this paper is to create a sentiment analysis

system that can tackle these challenges effectively.

By categorizing YouTube comments based on their

sentiment, such as positive or negative, and

identifying different types of sentences, the system

can offer content creators more detailed and useful

930

Raigar, D., Chapke, P., Dangare, P., Dhagude, P. and Malode, V.

Enhancing YouTube Comment Insights: A Machine Learning Approach to Sentiment Analysis.

DOI: 10.5220/0013735100004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 3, pages 930-933

ISBN: 978-989-758-763-4

feedback. With this information, creators can adapt

their content, better engage their audience, and

improve the overall quality of their videos.

Ultimately, this system aims to make it easier for

video creators to understand their audience’s

emotions and reactions, helping them stay relevant

and successful on the platform.

2 RELATED WORK

Pokharel, Rhitabrat, and Dixit Bhatta (2021).”

Classifying YouTube comments based on sentiment

and type of sentence.” This study explores emotion

classification of Indonesian YouTube comments

using various word embeddings and machine learning

models. The CNN method showed the best

performance, with an accuracy of 76.2% (Wei and

Zhang, 2024). Wei, Zhongliang, and Shunxiang

Zhang (2024).” A structured sentiment analysis

dataset based on public comments from various

domains.” This paper introduces a dataset offering

diversity and quality, enabling targeted analysis of

sentiment classification models, especially in

domain-specific contexts (Kumari, Anupriya, et al. ,

2024). Ruchita Kumari, et al. (2024).” Comment

Analyzer for Sentiment Analysis in Social Media and

E-Commerce Platforms.” The authors focus on

sentiment analysis in social media and e-commerce

platforms. Their model effectively classifies

comments by sentiment using NLP and machine

learning algorithms (Neve, Pachpute, et al. , 2024).

Sainath Pichad, et al. (2023).” Analysing Sentiments

for YouTube Comments using Machine Learning.”

This study uses Naive Bayes and SVM to classify

sentiments in YouTube comments (Cunha, Costa, et

al. , 2019). Singh, R., & Tiwari, A. (2021).”

Sentiment Analysis of YouTube Comments.” This

paper presents a sentiment analysis model that

classifies YouTube comments using SVM and

Decision Trees. Aditya Neve, Kalpesh Pachpute,

Bhimashankar Mathapati, Prerana Thorve (2024).”

YouTube Comment Sentiment Analysis.” This study

explores sentiment analysis for YouTube comments

using machine learning techniques, improving

engagement through classification methods

(Muhammad, Bukhori, et al. , 2019. Al Hujaili,

Rawan Fahad, and Wael MS Yafooz (2021).”

Sentiment analysis for YouTube videos with user

comments.” This paper presents a sentiment analysis

method for YouTube videos, focusing on user

feedback and employing deep learning algorithms for

analysis (Alberto, Lochter, et al., 2021). Muhammad,

Abbi Nizar, Saiful Bukhori, and Priza Pandunata

(2019). ” Sentiment analysis of positive and

negative YouTube comments using NB-SVM

classifier.” The paper discusses the hybrid Naive

Bayes-SVM classifier for sentiment classification,

yielding improved performance for polarity-based

classification (AlHujaili, Yafooz, et al. , 2021).

Bhuiyan, Hanif, et al. (2017).” Retrieving YouTube

video by sentiment analysis on user comment.”

(Bhuiyan, et al. , 2017), this study focuses on

retrieving YouTube videos based on sentiment

classification of user comments, emphasizing

machine learning models for extracting valuable

content (Sungheetha and Sharma, 2020).

3 RELATED GAP ANALYSIS

Several Gaps in Current Research Remain

Unaddressed

1. Domain-Specific Sentiment Analysis

Existing sentiment lexicons often struggle

with domain-specific language (Pichad,

Kamble, et al. , 2023).

2. Handling Informal Language and

Nuances Informal language and slang

hinder accurate sentiment classification

(Pokharel, Bhatta, et al. , 2021).

3. Big Data Challenges The growing volume

of online data presents challenges in

storage and analysis (Pichad, Kamble, et

al. , 2023).

4. Multimodal Sentiment Analysis

Sentiment can be expressed through text,

emojis, images, and videos(Alberto,

Lochter, et al. , 2021).

5. Real-Time Sentiment Analysis Research

into efficient and scalable real-time

sentiment analysis techniques is essential

(Singh and Tiwari, 2021).

4 MODELS

1) Rule-Based VADER

2) Classical ML Naive Bayes, SVM, Logistic

Regression, Random Forest

3) Deep Learning CNN, RNN, LSTM, BERT

Enhancing YouTube Comment Insights: A Machine Learning Approach to Sentiment Analysis

931

5 METHODOLOGIES

Our Approach Involves the Following Steps

5.1 Data Collection

YouTube comments are collected using YouTube’s

API.

5.2 Data Preprocessing

Tokenization, stop-word removal, stemming, and

lemmatization are applied.

5.3 Feature Extraction

Word embeddings (Word2Vec or GloVe) and TF-

IDF are used.

5.4 Modelling

Machine learning models (Naive Bayes, SVM, CNN)

classify comments by sentiment and sentence type.

5.5 Evaluation

Cross-validation and F1 score are used to evaluate

model performance.

6 APPLICATIONS

6.1 Content Creation

Helps YouTubers identify valuable feedback

6.2 Marketing and Analytics

Assists brands in gauging public sentiment

6.3 Audience Engagement

Enables more meaningful viewer engagement

6.4 Social Media Analysis

Provides insights into trends and opinions

6.5 Spam Detection

Enhances spam filtering capabilities

Figure 1. The process of collecting data,

preprocessing it, and utilizing classifier models

Figure 2. System Architecture

7 CONCLUSIONS

As YouTube channels grow, they receive large

volumes of comments that provide valuable

feedback, crucial for understanding audience

sentiment and improving engagement. However,

existing sentiment analysis approaches focus only

on single-language positive/negative classification,

and struggle with the informal language in YouTube

comments. This project addresses these limitations

by building a system to classify comments based on

sentiment (positive, negative, neutral) and sentence

types, using advanced NLP techniques and machine

learning models. Our system will support

multilingual analysis, increasing accessibility, and

its performance will be evaluated using cross-

validation and F1 scores to help creators improve

their content.

INCOFT 2025 - International Conference on Futuristic Technology

932

REFERENCES

Alberto, T. C., Lochter, J. V., & Almeida, T. A. (2021).

Tubespam: Comment spam filtering on YouTube. In

2021 IEEE 14th International Conference on Machine

Learning and Applications (ICMLA). IEEE.

AlHujaili, R. F., & Yafooz, W. M. S. (2021). Sentiment

analysis for YouTube videos with user comments. In

2021 International Conference on Artificial

Intelligence and Smart Systems (ICAIS). IEEE.

Bhuiyan, H., et al. (2017). Retrieving YouTube video by

sentiment analysis on user comment. In 2017 IEEE

International Conference on Signal and Image

Processing Applications (ICSIPA). IEEE.

Cunha, A., Costa, M., & Pacheco, M. (2019). Sentiment

analysis of YouTube video comments using deep neural

networks. Springer, 11508.

https://doi.org/10.1007/978-3-030-20912-451

Kumari R., Anupriya, Singh, S., & Shukla, H. (2024).

Comment analyzer for sentiment analysis in social

media and e-commerce platforms. International

Journal of Innovative Research in Computer Science

Technology (IJIRCST), 12(Special Issue-1), March

2024.

Muhammad, A. N., Bukhori, S., & Pandunata, P. (2019).

Sentiment analysis of positive and negative YouTube

comments using Na"{i}ve Bayes–support vector

machine (NB-SVM) classifier. In 2019 International

Conference on Computer Science, Information

Technology, and Electrical Engineering (ICOMITEE).

IEEE.

Neve, A., Pachpute, K., Mathapati, B., & Thorve, P. (2024).

YouTube comment sentiment analysis. International

Journal of Creative Research Thoughts (IJCRT), 12(6),

June 2024.

Nawaz, S., Rizwan, M., & Rafiq, M. (2019).

Recommendation of effectiveness of YouTube video

contents by qualitative sentiment analysis of its

comments and replies. Pakistan Journal of Science,

71(4 Suppl.), 91-97.

Pichad, S., Kamble, S., Kalamb, R., & Chavan, S. (2023).

Analysing sentiments for YouTube comments using

machine learning. International Journal for Research

in Applied Science and Engineering Technology

(IJRASET), 11(V), May 2023.

Pokharel, G., & Bhatta, D. (2021). Classifying YouTube

comments based on sentiment and type of sentence.

arXiv preprint arXiv:2111.01908.

Savigny, J., & Purwarianti, A. (2017). Emotion

classification on YouTube comments using word

embedding. In 2017 International Conference on

Advanced Informatics, Concepts, Theory, and

Applications (ICAICTA). IEEE.

Singh, R., & Tiwari, A. (2021). Sentiment analysis of

YouTube comments. International Journal of Scientific

Research in Engineering and Management (IJSREM),

5(5), 1-5.

Sungheetha, A., & Sharma, R. (2020). TransCapsule model

for sentiment classification. Journal of Artificial

Intelligence, 2(3), 163-169.

Timani, H., Shah, P., & Joshi, M. (2019). Predicting

success of a movie from YouTube trailer comments

using sentiment analysis. In Proceedings of the 2019

6th International Conference on Computing for

Sustainable Global Development (INDIACom 2019)

(pp. 584-586).

Wei, Z., & Zhang, S. (2024). A structured sentiment

analysis dataset based on public comments from

various domains. Data in Brief, 53, 110232.

Enhancing YouTube Comment Insights: A Machine Learning Approach to Sentiment Analysis

933