Volume-3-Issue-6-June,2025 International Journal of Modern Science and Research Technology

ISSN NO-2584-2706

Sentiment Analysis on Text based Data of Social

Media using Deep Learning
Shruti Sudhakar Chavan; Dr S.S.Sherekar
Department Of Computer Science
Sant Gadge Baba Amravati University
Amravati Maharashtra India

Abstract – Sentiment Analysis Using Deep A. Sentiment Analysis

Learning Is The Process Of Automatically Sentiment Analysis Is Nothing But
Identifying And Categorizing Emotions Classifying Opinion Of People Into Negative ,
Expressed In Textual Data By Leveraging Positive, Neutral .Tweet Data Holds Valuable
Neutral Network Models That Learn Complex Insights That Can Guide Company Policy
Pattern In Language. This Paper Presents A Decisions Through Sentiment Analysis. This
Robust Framework For Sentiment Involes Using Text Mining And Natural
Classification Of Social Media Text Using A Language Processing To Extract And Classify
Bidirectional Long Short – Term Memory Sentiments From Tweets. [1] Public
(BiLSTM) Model . A BiLSTM Based Model Comments And Reviews From Platforms
Is Use To Analyse Social Media Data, With Like Twitter And Facebook Are Key For
Preprocessing, Class Balancing, And Feature Analysing Overall Sentiment On Products ,
Extraction Stages. The Model Achieves 95% Services, Or Global Events . [2]
Accuracy , Effectively Predicting Positive, Twitter Captures Public Emotions Through
Negative , And Neutral Sentiments , Short Posts, And Sentiment Analysis Interpret
Demonstrating Its Strength For Real-Time These Opinions On Event Or Policies [3].
Sentiment Monitoring . Sentiment Polarity In Tweets About The
Digital India Mission , Showing Promising
Keywords – Sentiment Analysis , Social And Outlining In Tweets About The Digital
Media, Deep Learning , Text Based Data India Mission , Showing Promising Results
And Outlining Plans To Enhance Accuracy
I.Introduction By Addressing Challenges Like Sarcasm ,
Sentimental Analysis Refers To The Task Of Negation, And Emotions[4] Sentiment-
Determining The Sentimental Polarity – Specific Word Embedding Using Modifies
Positive , Negative , Or Neutral – Expressed Delta Tf-Idf To Enhance Twitter Sentiment
Within A Given Block Of Block Of Text. Analysis , With Future Work Focused On
Deep Learning , A Subfield Of Machine Capturing Word Order.[5]
Learning , Utilizes Artificial Neutral Network
Architectures To Model Complex Pattern In B. Deep Learning
Data . In This Study , A Deep Learning – Deep Learning Is A Subset Of Ai Which Use
Based Approach Employing A Bidirectional Artificial Neural Networks To Learn From
Long Short – Term Memory (Bilstm) Data. Cnn- Based Deep Learning Model For
Network Is Utilized To Train On Labeled Classifying Tweet Sentiments Into Positive,
Text Data And Perform Sentimental Negative, And Neutral , Achieving 63.59%
Classification, Effectively Predicting Whether Test Accuracy Using Preprocessed Tweets
The Sentiment Is Positive , Negative , Or And Fasttext Word Embedding. [6]
Neutral
IJMSRT25JUN32 www.ijmsrt.com 185
DOI: https://doi.org/10.5281/zenodo.15637232
Volume-3-Issue-6-June,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

C. Sentiment Analysis On Text Based Data Without Violating Oublic Sentiment. The
Of Social Media Using Deep Learning Study
The Text Is Preprocessed By Cleaning And Employs Machine Learning Classifiers,
Tokenizing , Then Analysed Using The Polarity-Based Analysis, And Deep Learning
Labelled Semeval Tweet Dataset With Model To Classify Tweets As Positive Or
Sentiments Categorized As Negative, Positive Negative , Aiming To Handle The Diverse
, Neutral.[7] Range Of Opinions And Enabling Real Time
Different Models Are Use After That [8]. We Sentiment Classification.
Can Also Do Political Analysis By Using Prasanna Kumar Et.Al (2024) [20] This Paper
Sentiment Analysis [9].After Training Model Aim To Improve Sentiment Analysis On
We Can Do Final Sentiment Prediction Social Media By Developing A Custom
Weather It Is Positive, Negative, Neutral.[10] Framework(Smsa) That Include A Specially
Created Emotion Thesaurus. It Explores How
II.Related Work Different Deep Learning Model Perform
Hong Soon Goo Et.Al(2017)[22] The Paper Under Varied Settings And Highlights A
Emphasizes The Growing Popularity Of Novel Approach To Building Sentiment
Social Media And Its Potential For Tasks Dictionaries . The Study Shows Improved
Like Prediction And Sentiment Analysis , Accuracy And Insights Compared To Earlier
Particularly Using Platforms Like Twitter. Methods, Offering Promising Directions For
Due To The Massive Volume Of Future Applications
Unstructured Data , Machine Learning – Dilesh Tanna Et.Al(2020)[26] This Paper
Specially Deep Learning With Deep Introduces A Custom Social Media Platform
Feedforward Neutal Networks – Is Essential With Built-In Sentiments Analysis To Access
For Effective Analysis , Achieving Around User Activity. It Rates Sentiments, Generates
75% Accuracy In Experimental Results. Reports For Admins , And Helps Identity
Qurat Tul Ain Et.Ai(2017) [23] The Paper Issues Like Emotional Distress. The Platform
Highlights The Vast Amount Of User Also Personalizes Content And Supports
Generated Data On Web – Like Opinions And Sharing Across Other Social Media.
Emotions On Social Network, Forums , And
Blogs – And The Growing Importance Of IV. Litreture Survey
Sentiments Analysis In Organizing This Wadzani Et.Al [11] This Give A Detail
Unstructured Data . It Notes The Challenge Overview Of The Use Of Machine Learning
Of Limited Labelled Data In Nlp And And Deep Learning Model In Detecting
Emphasizes The Integration Of Deep Depression On Social Media . The Deep
Learning Techniques, Such As Deep Neural Learning Model Turn Out To Be More
Network And Cnns, For Effective Sentiment Realistic And Gives Better Accuracy, In This
Classification And Addressing Issues Like Paper Cnn,Lstm And Rnn Model Are Used
Cross – Lingual Analysis .[11]
Yogesh Chandra Et.Al [24] The Paper Aditya Patil Et.Al(2023)[12] In This Research
Emphasizes The Need For Sentiment Paper An Approach Is Used To Enhance
Analysis Of User Generated Data On Social Sentiment Analysis On Social Media Data
Media To Aid Decision Making For Through Text Analysis An Predictive
Government And Defense Organization , Modelling [12]
Especially During Critical National Event . Kottala Sri Yogi Et.Al[13] This Research
By Analyxing Opinions Shared On Platforms Explores The Use Of Machine Learning ,
Like Twitter And Facebook , Insight Can Be Particularly Nb,Svm And Deep Learning
Gained Model Like Cnns And Rnns, For Sentiment
Analysis On Social Media. It Highlights Deep
Learning’s Superior Accuracy And Ability To
IJMSRT25JUN32 www.ijmsrt.com 186
DOI: https://doi.org/10.5281/zenodo.15637232
Volume-3-Issue-6-June,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

Capture Subtle Emotions In Text , Despite Mohammad Et.Al [21] This Paper Introduces
Challenges In Real Time Processing. [13] A Hybrid System Combining Text Mining
Israt Jahan Et.Al (2024) [14] This Research And Neural Networks For Sentiment
Shows Random Forest Outperforms Other Classification , Comparing It With Machine
Models In Predicting Heart Attacks Using Learning And Deep Learning Algorithm
Patient And Ecg Data . It Emphasizes The Using A Dataset Of 1 Million Tweets, The
Importance Of Advanced Machine Learning System Achieved An Accuracy Rate Of
In Improving Early Diagnosis And Patient 83.7% Outperforming Standard Supervised
Outcomes. [14] Approaches [21]
Youngchaul Et.Al [15] This Study
Demonstrated That Incorporating Sentiment V. Proposed Method
Polarities From Twitter Data Into Predictive The Design Of This Proposed Method Is
Model Like Lstm And Gru Improves The Illustrated In Figure 1. The Collected
Accuracy Of Forecasting Confirmed Covid- Database Of Social Media Text Is Labelled
19 Cases . It Highlights The Value Of With Sentiments Like Positive ,Negative ,
Qualitative Opinion Mining For Epidemic Neutral . The Data Will Go Through
Prediction And Suggests Future Research Preprocessing Stages . Then The Next Stage
Should Expand Data Sources And Explore Is Handling Class Imbalance . The Next Stage
Additional Modelling Approaches. [15] Is Feature Extraction . The Bilstm Deep
Sashank Saya [16] This Flask – React App Learning Model Is Used In Training Phase .
Performs Real – Time Sentiment Analysis On The Final Stage Is Sentiment Prediction
Text And Images Using Machine Learning ,
Achieving 82% Accuracy With Gradient
Boosting And Vgg 16 , Supporting Use Cases
Like Social Media Monitoring And Content
Moderation . [16]
Mr Swapnil Et.Al [17] This Study Analyses
Political Youtube Comments On Bjp And Inc
Using The Afinn Lexicon And Various
Machine Learning Models, Finding That Svm
And Random Forest Offer The Best [26] - Bilstm Structure
Performance For Sentiment Analysis , While
The Afinn Lexicon And K-Nn Showed Lower
Accuracy . DATA COLLLECTION
Sunasi Et.Al [18] This Paper Present An
Ensemble Method Using Random Forest , K –
Nearest Neighbor , Naïve Bayes And Svm To DATA CLEANING AND
Detect Sarcasm And Its Types In Social PREPROCESSING
Media , Improving Sentiment Analysis
Accuracy With Proven Reliability [18].
Nabaa Et.Al [19] This Study Explores HANDLING CLASS IMBALANCE
Statistical Challenges In Social Media Data
Analysis , Focusing On Sentiment Tracking .
By Using Advanced Statistical Techniques FEATURE EXTRACTION
Like Nlp And Machine Learning , It BiLSTM(DEEP LEARNING MODEL)
Addresses Issues Such As Data Sparsity ,High
Dimentionality, And Noise , Improving
Sentiment Classification Accuracy By 15% FINAL SENTIMENT PREDICTION
And Reducing Noise By 20%. [19].

IJMSRT25JUN32 www.ijmsrt.com 187

DOI: https://doi.org/10.5281/zenodo.15637232
Volume-3-Issue-6-June,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

Fig 1 : Overview Of Model  Assign Higher Weights In Minority

Classes During Odel Training To Ensure
A. Data Collection Balanced Learning
The Data Set Utilized In This Research
Contains Text Used For Differ Social Media C. Feature Extraction
Platform Such As Twitter , Instagram It Starts With Tokenization Which Convert
,Telegram , Whatsapp , Telegram ,Quora . We Text Into Numerical Tokens That Model Can
Will Give Sentiment Label To This Text Process , Then There Is Need For Padding In
,Then By Using Deep Learning Model We Which Neural Network Required Input
Will Do Sentiment Classification To Get Our Sequences Of Equal Length.
Desired Output .
D.Deep Learning Model
B. Data Cleaning and Preprocessing The System Uses Bidirectional Long Short
Data Cleaning And Preprocessing Is Essential Term Memory (Bilstm) Network , Ideal For
To Transform Raw Text Into A Format Sequential Data Like Text Because It
Suitable For Deeo Learning Model. It Starts Captures Both Past And Future Context In
With Text Normalization Which Convert All Sentences .
Text To Lowercase For Consistency And
Remove Extra Whitespace , Table And Model Architecture
Newline Charecters, Then Noise Removal
For Eliminate Url, Hashtags, Mentions, EMBEDDING LAYER
Special Character, And Emoji And Remove
Numbers Unless They Carry Meaningful Bidirectional-LSTM LAYERS
Sentiments. Tokenization Split Text Into
DROPOUT LAYERS
Individual Words(Tokens) For Easier
Processing And Stopword Removal Filter DENSE LAYER
Out Comman Words (E.G., “The”,”Is”,”In”)
That Do Not Contribute Significantly To
Sentiment Analysis, Retain Negative Words SOFTMAX ACTIVATION FUNCTION
(E.G , “Not”,”Never”,”Isn’t”) As They
Influence Sentiment Meaning. Finally
Stemming Reduce Words To Their Root Fig 2. : Deep Learning Model Architecture
Form (E.G “Running””Run”) To
Standardize Different Forms Of Same Word
 Embedding Layer
C. Handling Class Imbalance Convert Tokens Into Dense Vectors,
In Sentiment Datasets, Certain Classes (Like Representing Semantic Relationships
Positive ) Often Dominate . This Imbalance  Bi-Directional LSTM Layers:
Can Cause The Model To Favour The Processes Text In Both Forward And
Majority Class Backward Directions To Capture Context
Techniques To Address Imbalance : Especially Effective For Understanding
 Resampling : Sentences-Level Meanings
 Oversampling : Increase Samples Of  Dropout Layers:
Minority Classes Randomly Deactivates Neurons Training To
 Undersampling : Reduce Samples Of Prevent Overfitting
Majority Class  Dense Layers:
 Data Augmentation Learn Complex Patterns And Relationships In
 Generate Synthetic Data Using The Data .
Paraphrasing Or Translation Techniques Softmax Activation Function :
 Class Weight Adjustment :
IJMSRT25JUN32 www.ijmsrt.com 188
DOI: https://doi.org/10.5281/zenodo.15637232
 Volume-3-Issue-6-June,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

Outputs Probabilities For Each Sentiment Fig 5 : F1-Score

Class
The Class With The Highest Probability Is
Chosen As The Final Prediction
In Model Training , The Loss Function Is
Used And Adam Optimizer Combines
Benefits Of Adagrad And Rmsprop , Adjusts
Learning Rates Dynamically For Faster
Convergence. The Regularization Techniques
Suh As Early Stopping And Dropout
Regularization Is Used In Model Training. For The Above Table We Understand That
D. Final Sentiment Prediction Fig (3) Shows Precision Of Positive
Final Sentiments Predictions Reveals Weather Sentiments Is 93 % , Negative Is 95 % And
The Text Is Positive , Negative Or Neutral.
For Neutral Is 97 % , The Overall Precision Is
VI. Result and Discussion 95 % . The Fig (4) Shows Recall Is 96 % For
In This Section We Will Discuss The Positive Sentiments , 98 % For Negative
Result Of Our Model That Is Performance Sentiments And 92 % For Neutral Sentiments
Metric , Therefore The Overall Recall Is 95.33 %.

Table 1 : Result Obtained Of Proposed Class Positiv Negativ Neutra Accurac

e e l y
Model
Precisio 0.93 0.95 0.97 0.95
n
Recall 0.96 0.98 0.92 0,95
F1 – 0.94 0.96 0.94 0.95
Score
Support 6665.0 6715.0 6620.0 1.0

Fig (4) Shows F1- Score Tells The Overall

Performance Of Model , So For Positive
Sentiments It Is 94 % , Negative It Is 96%
And Neutral It Is 94% And Therefore Overall
Fig 3 : Precision
Performance Of This Model Is 95 % .
The Overall Accuracy Is 95 % .

VII. Conclusion
To Ensure Consistency And Reduce Noise ,
All Text Was Lowercased , Punctuation
Removed And Stopwords Filtered Out –
Except For Crucial Negations Like “Not ”
Fig 4 : Recall And “Never” , Which Significantly Affect
Sentiment. To Address Sentiment Imbalance
In Social Media Datat, Resampling
Techniques Balanced The Dataset Across
Positive, Negative, Neutral Classes. The
Model Used A Bidirectional LSTM To
Capture Context In Both Directions, With
Dropout Layers To Prevent Overfitting And
A Softmax Layer For Sentiment
Classification, Performance Was Evaluated
IJMSRT25JUN32 www.ijmsrt.com 189
DOI: https://doi.org/10.5281/zenodo.15637232
Volume-3-Issue-6-June,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

Using Precision , Recall , F1 – Score Ad Sentiment Predictions, And The Model

Support , With Early Stopping And Achieved Near – Perfect Accuracy With
Checkpoints Optimizing Training . An Balanced , Reliable Results.
Interactive Loop Enabled Real- Time
Fig 6 : Support
IEEE International Conference On Big Data
References (Big Data) 2016
[1] Kusrini,Machamad Mashuri,"Sentiment [9] Gaurav Dubey,Silpi Chawla,Kirandeep
Analysis In Twitter Using Lexicon Based Kaur,"Social Media Analysis Of Indian
And Polarity Multiplication",IEEE,2019 Political Diplomat ",7th International
[2] Huma Parveen,Prof. Shikha Conference On Cloud Computing, Data
Panday,"Sentiment Analysis On Twitter Data- Science & Engineering – Confluence 2017
Set UsingNaive Bayes [10] Zhaoxia WANG, Chee Seng CHONG,
Algorithm",IEEE,2016 Landy LAN, Yinping YANG, Seng Beng HO
[3] Rupali Patil ,Nishat Gada,Krisha Gala And Joo Chuan TONG,"Fine-Grained
,"Twitter Data Visualization And Sentiment Sentiment Analysis Of Social Media With
Analysis Of Article 370",IEEE Emotion Sensing"Future Technologies
[4] Mishra, P., Rajnish, R., & Kumar, P. Conference 2016
(2016). Sentiment Analysis Of Twitter Data: [11] Adyan Marendra Ramadhani,Hong Soon
Case Study On Digital India. Amity Goo,"Twitter Sentiment Analysis Using Deep
University, Lucknow, India. Learning Methods",7th International Annual
[5] Rania Othman,1 Youcef Abdelsadek,2 Engineering Seminar (Inaes), Yogyakarta,
Kamel Chelghoum,2 Imed Kacem,2 Rim Indonesia ,2017.
Faiz3,"Improving Sentiment Analysis In [12] Q. Tul Ain, M. Ali , A Riaz , A Noureen
Twitterusing Sentiment Specific Word , M Kamran , B Hayat And A. Rehman”
Embeddings", 10th IEEE International Sentiment Analysis Using Deep Learning
Conference On Intelligent Data Acquisition Techniques: A Review” (IJACSA)
And Advanced Computing Systems: International Journal Of Advanced Computer
Technology And Applications18-21 Science And Applications, Vol. 8, No. 6,
September, 2019 2017.
[6] Pota, M., Esposito, M., Palomino, M. A., [13] Yogesh Chandra,Antoreep
& Masala, G. L. (2018). A Subword-Based Jana,"Sentiment Analysis Using Machine
Deep Learning Approach For Sentiment Learning And Deep Learning",7th
Analysis Of Political Tweets. 2018 32nd International Conference On Computing For
International Conference On Advanced Sustainable Global Development
Information Networking And Applications (Indiacom),2020
Workshops (WAINA), 604–609. [14] Prasanna Kumar Rangarjan,Bharathi
Https://Doi.Org/10.1109/WAINA.2018.00147 MohanGurusamy,Gayathri Muthurasu,Rithani
[7] Rania Othman,1 Youcef Abdelsadek,2 Mohan,GundalaPallavi,Sulochana
Kamel Chelghoum,2 Imed Kacem,2 Rim Vijayakumar,Ali Altalbe,"The Social Media
Faiz3,"Improving Sentiment Analysis In Sentiment Analysis Framework: Deep
Twitterusing Sentiment Specific Word Learning For Sentiment Analysis On Social
Embeddings", 10th IEEE International Media",International Journal Of Electrical
Conference On Intelligent Data Acquisition And Computer Engineering (IJECE) Vol. 14,
And Advanced Computing Systems: No. 3, June 2024
Technology And Applications18-21 [15] Tanna, D., Dudhane, M., Sardar, A.,
September, 2019 Deshpande, K., & Deshmukh, N. (2020).
[8] Parul Sharma ,Teng-Sheng Moh Sentiment Analysis On Social Media For
“Prediction Of Indian Election Using Emotion Classification. In Proceedings Of
Sentiment Analysis On Hindi Twitter” The International Conference On Intelligent
IJMSRT25JUN32 www.ijmsrt.com 190
DOI: https://doi.org/10.5281/zenodo.15637232
Volume-3-Issue-6-June,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

Computing And Control Systems (ICICCS Https://Doi.Org/10.1371/Journal.Pone.030984

2020) (Pp. 233–237). IEEE. 2
Https://Doi.Org/10.1109/ICICCS48265.2020. [21] Sashank Saya,Shreelaxmi K
9161060 Malawade ,Smaranya Vijaya
[16] Wadzani Aduwamai Gadzama , Danlami Krishna ,Vaishnavi K S ,Shylaja
Gabi , Musa Sule Argungu , Hassan Umar B ,"Sentiment Analysis For Social Media
Suru,"The Use Of Machine Learning And Presence",Research Square ,2024.
Deep Learning Models In Detecting [22] Goje, S. P., & Patil, R. H. (2024).
Depression On Social Media: A Systematic Sentiment Analysis Of Political Parties On
Literature Review", Social Media: A Machine Learning And
Www.Sciencedirect.Com/Journal/Personalize Lexicon-Based Approach. International
d-Medicine-In-Psychiatry,2024 Journal Of Scientific Research In Computer
[17] Aditya Patil, Swati Patil, Prajakta Science, Engineering And Information
Patil,Dr.Vaishali Patila, Sanket Technology, 10(5), 233–246.
Lodhac,Dr.Sudhir Chitnisb,Dr. Arun Https://Doi.Org/10.32628/CSEIT24105103
Patilb,"An Approach To Enhance Sentiment [23] Sunusi Kabir Alaramma1, Adamu
Analysis On Social Media Data Through Text Adamu Habu2, Badamasi Imamu Ya’u1, And
Analytics And Predictive Modeling",Research Abdullahi Gamsha Madaki,"Sentiment
Gate , Empirical Economics Letters, 22 (2): Analysis Of Sarcasm Detection In Social
(August 2023) Media",Gadau Journal Pure Alli Sci, 2(1): 76-
[18] Kottala Sri Yogi,Dankan Gowda 82 (2023)
V,Galiveeti Poornima,KDV [24] Nabaa Muhammad Diaa1, Saba Sabah
Prasad,Srinivas.D,P Vishnu Ahmed2, Hayder Mahmood Salman3, Wafaa
Prasanth,"Comparative Analysis Of Machine Adnan Sajid4,"Statistical Challenges In Social
Learning Techniques For Detecting Media Data Analysis Sentiment Tracking And
Sentiments In Social Media",Research Gate Beyond",Journal Of Ecohumanism 2024
,May 2024. [25] Mohammed H. Abd El-Jawad,Rania
[19] Israt Jahan, Md Nakibul Islam , Md Hodhod,Yasser M. K. Omar,"Sentiment
Mahadi Hasan And Md Rafiuddin Siddiky, Analysis Of Social Media Networksusing
"Comparative Analysis Of Machine Learning Machine Learning"
Algorithms For Sentiment Classification In [26] Wang, H., Wang, J., Cao, L., Li, Y., Sun,
Social Media Text", World Journal Of Q., & Wang, J. (2021). A Stock Closing Price
Advanced Research And Reviews, 2024. Prediction Model Based On CNN-Bislstm.
[20] Song Y, Yoon B. Prediction Of Complexity, 2021, 5360828.
Infectious Diseases Using Sentiment Analysis Https://Doi.Org/10.1155/2021/5360828
On Social Media Data. Plos ONE.
2024;19(9):E0309842.

IJMSRT25JUN32 www.ijmsrt.com 191

DOI: https://doi.org/10.5281/zenodo.15637232