LEVERAGING MACHINE

LEARNING
FOR EMOTION DETECTION
AND OPINION MINING
By Project Guide:
Navuru Sahithya- 160122737153 Mr. B. Harish Goud
Sai Tejaswi Edara-160122737159
Snikitha Govindu-160122737161
 1. Abstract

2. Introduction

3. Literature Survey

4. Problem Statement
Agenda
5. Methodology

6. Results

7. Conclusion

8. References
Abstract
• Sentiment analysis is a branch of natural language processing that
focuses on identifying and analyzing emotions and opinions in text. Its
main goal is to determine whether expressions are positive, negative, or
neutral.
• This field is vital for understanding subjective opinions from sources like
social media, blogs, and reviews, acting as a gauge for public sentiment.
By using machine learning algorithms, sentiment analysis processes text
autonomously, eliminating the need for manual evaluation.
• However, it faces challenges such as language nuances, cultural
differences, and evolving sentiment expressions, requiring continuous
improvement to remain effective.
Introduction
• Sentiment Analysis (SA) in NLP is vital for gauging attitudes and emotions in
text. It utilizes machine learning to automate analysis, extracting insights
about sentiments towards various elements or products. By training models
with annotated emotional text examples, SA algorithms accurately discern
sentiments, considering context, sarcasm, and nuanced language.

• SA operates at different levels: document level for overall sentiment,

sentence level for individual analysis, and sub-sentence level for nuanced
understanding.

• In essence, SA is a potent tool for exploring sentiments in textual data,

aiding organizations in decision-making and improving outcomes.
 Literature Survey
Sentiment Analysis This research utilizes Amazon customer review data to discern
using Machine Learning positivity, negativity, and neutrality, employing machine learning
Technique; A Literature techniques. It aims to extract insights from customer feedback,
Survey (December illuminating sentiments towards various products and services on
2020, Journal) Amazon. Another study investigates Alzheimer's Disease stigma on
Twitter amidst the COVID-19 pandemic. It examines the stigma
surrounding Alzheimer's disease on Twitter, using machine
learning techniques to analyze tweets.
Application of Machine -Leveraging advanced machine learning algorithms, researchers
Learning for Sentiment aim to dissect the sentiment encapsulated within movie reviews
Analysis of Movies on IMDB, thereby offering invaluable insights into audience
Using IMDB Rating perceptions of diverse films.
(2020,Conference) - The study meticulously evaluates the accuracy of various
machine learning models, shedding light on the effectiveness of
the proposed model vis-à-vis existing approaches in achieving
precision and reliability.
5
 Literature Survey
Sentiment Analysis on -Next paper talks about Twitter, as a dynamic microblogging
Tweets Using Machine platform, presents a rich tapestry of data ripe for analysis of
Learning and prevailing trends and sentiments on a macroscopic scale.
Combinational Fusion
(2019, Conference) -This paper introduces a novel two-stage data analytic
approach tailored for dissecting the intricacies of natural
language processing (NLP) and sentiment analysis labeling on
Twitter data.

- Through the adept utilization of machine learning

techniques and innovative combinatorial fusion strategies, the
authors aim to extract profound insights from tweets, thereby
contributing significantly to a deeper comprehension of public
opinion and sentiment trends across social media platforms.
Problem Statement
• Growing complexity and volume of user-generated content on digital
platforms.
• Difficulty in extracting and interpreting meaningful insights from raw
text.
• Traditional data analysis methods often miss nuanced expressions
such as sarcasm, cultural references, and evolving language.
• Need for scalable solutions to efficiently and accurately process large
datasets without extensive manual oversight.

7
Methodology
• Start: This marks the start of our sentiment analysis journey, where we
aim to understand and categorize the sentiment in textual data,
forming the foundation of our analytical endeavor.

• Data Collection: We gather essential data for sentiment analysis by

sourcing diverse text samples, like customer reviews and social media
posts, along with their sentiment labels indicating positivity,
negativity, or neutrality.

• Data Preprocessing: Before data enters our models, we preprocess it.

This involves cleaning by removing noise like punctuation,
standardizing text through lowercasing, tokenization, and eliminating
stop words.

8
Methodology
• Sentiment Identification and Classification using Learned Classifiers: Our
sentiment analysis method focuses on training machine learning
classifiers, such as Support Vector Machines, Naive Bayes, and neural
networks. They rigorously learn patterns from preprocessed data,
enabling accurate sentiment predictions for new text samples.

• Analysis and Evaluation of Models: Subsequent to the training phase, we

embark on the critical task of evaluating the performance of our trained
classifiers.

• End: This marks the culmination of our sentiment analysis journey,

wherein we have meticulously developed and evaluated machine
learning models proficient in identifying and categorizing sentiment
within textual data, paving the way for actionable insights and informed
decision-making.
Results

10
Results
- Initial observation: Positive sentiments were predominant over negative and
neutral sentiments through graphical data visualization.
- N-Gram analysis: Performed to scrutinize unigrams, bigrams, and trigrams
associated with each sentiment category for refining understanding of textual
nuances and improving initial results accuracy.
- Dataset balancing: Converted sentiments into numerical values and utilized
TF-IDF Vectorizer for feature extraction to ensure dataset balance.
- Resampling: Imbalanced data underwent resampling before being split into a
75/25 ratio for training and testing.
- K-Fold Cross Validation: Employed on the pre-resampled dataset to validate
results thoroughly and benefit from its robustness against data imbalance.

11
Result
- Core phase: Developed sentiment analysis models using machine learning
techniques.
- Tested models: K-Nearest Neighbors (KNN), Support Vector Classifier (SVC),
Logistic Regression, Random Forest, and Decision Tree Classifiers.
- Evaluation metrics: Test accuracy, precision, and recall.
- Results: Logistic Regression outperformed others slightly in accuracy
compared to SVC during 10-Fold Cross Validation.
- Fine-tuning: Optimized Logistic Regression model using Grid Search for
optimal hyperparameters.
- Training set accuracy: 94.80%.
- Test set accuracy: Impressive 95.21%.
- F1 Score: Achieved 95% across all sentiment categories.
- Conclusion: Model demonstrated effectiveness through comprehensive
approach from data visualization to rigorous testing and tuning.
12
Conclusion
- Project overview: Comprehensive process of managing, processing, and analyzing vast
textual data to categorize sentiments across digital platforms.
- Utility: Illustrates broad applicability in market research, brand monitoring, social
media analysis, and political discourse.
- TF-IDF method: Strategically tailored for nuanced analysis, accurately capturing term
significance within documents to enhance overall analysis.
- Decision factors: Size of dataset, required accuracy level, and available resources
guided the choice of TF-IDF.
- Insights: Provide stakeholders with clearer understanding of public opinion and
sentiment trends.
- Implications: Aid strategic decision-making in marketing, product development,
customer service, and policy formulation, facilitating dynamic adaptation to consumer
needs and market changes.
References
• [1] Ali Athar; Sikandar Ali; Muhammad Mohsan Sheeraz; Subrata Bhattachariee; Hee-Cheol
Kim“Sentimental Analysis of Movie Reviews using Soft Voting Ensemble-based Machine
Learning“Publisher: IEEE,Link:IEEE xplorer.
• [2] G Prema Arokia Mary; M S Hema; R Maheshprabhu; M Nageswara Guptha. “Sentimental Analysis of
Twitter Data using Machine Learning Algorithms”.Publisher: IEEE,Link:IEEE xplorer.
• [3] Muskan Agarwal; Richa Goyal; Eshika Verma; Hemlata Goyal; Gulrej Ahmed; Sunita
Singhal.”Predictive Sentimental Analysis of Spam Detection using Machine
Learning“,Publisher:IEEE,Link: IEEE xplorer.
• [4] Nirag T. Bhatt1, Asst. Prof. Saket J. Swarndeep2,” Sentiment Analysis using Machine Learning
Technique: A Literature Survey”, Publisher: irjet,Link:www.irjet.net
• [5] P Ancy Grana,” Sentiment analysis of text using machine learning Publisher: International
Research Journal of Modernization in Engineering Technology and
Science,Link:https://www.irjmets.com/.
• [6] Kaggle website for the CSV file - https://www.kaggle.com/eswarchandt/amazon-music-reviews?
select=Musical_instruments_reviews.csv