Sentiment Analysis

Ms. Pritam Dahiphale Dr. Snehal Rathi Mr. Sujal Sunil Chavan
Department of Computer Department of Computer Department of Computer
Engineering Engineering Engineering
Vishwakarma Institiute of Vishwakarma Institiute of Vishwakarma Institiute of
Information Technology Information Technology Information Technology
Pune, India Pune, India Pune, India
pritam.22220044@viit.ac.in snehal.rathi@viit.ac.in sujal.22220104@viit.ac.in
Mr. Nishant Mulane Mr. Kaushal Khachane Mr. Harshvardhan Sasane
Department of Computer Department of Computer Department of Computer
Engineering Engineering Engineering
Vishwakarma Institiute of Vishwakarma Institiute of Vishwakarma Institiute of
Information Technology Information Technology Information Technology
Pune, India Pune, India Pune, India
nishant.22110421@viit.ac.in kaushal.22220110@viit.ac.in harshvardhan.22110393@viit.ac.in
Abstract- Sentiment analysis, also known as opinion domains such as social media monitoring, customer
mining, is a technique used to determine the emotional tone feedback analysis, market research, and brand management,
behind a piece of text. It involves analyzing and allowing businesses to make data-driven decisions and
categorizing the opinions expressed in the text as positive, formulate effective strategies to enhance their reputation and
negative, or neutral. Sentiment analysis has gained improve customer satisfaction.
significant attention in recent years due to the explosion of
online content and the need for businesses to understand and I.2 PURPOSE AND SCOPE
respond to customer feedback.
The scope and purpose of sentiment analysis is broad
and diverse, making it applicable to various domains.
The purpose of sentiment analysis is to gain insights into
Let's explore some of the key areas where sentiment
public opinion, customer satisfaction, brand reputation, and
market trends. By analyzing the sentiment of customer analysis is utilized:
reviews, social media posts, and other text data, businesses
can make informed decisions, improve customer experience, Marketing: Sentiment analysis plays a vital role in
and identify potential issues before they escalate. understanding customer perception of a brand,
product, or advertising campaign. By analyzing the
Sentiment analysis can be performed using various methods sentiment of customer reviews, social media posts,
and techniques, including natural language processing and other text data, marketers can gain insights into
(NLP), machine learning, and statistical analysis. NLP what resonates with their target audience. This
algorithms analyze the semantic and syntactic structure of information helps them tailor their messaging
text to extract meaning and sentiment. Machine learning accordingly, enhancing brand positioning and
models are trained on labeled data to classify new text improving customer engagement.
inputs into positive, negative, or neutral sentiments.
Statistical analysis involves measuring the frequency and
Customer Service: In the realm of customer service,
distribution of sentiment words and phrases in a given text.
sentiment analysis is a valuable tool for categorizing
Keywords: Sentiment analysis, ML, NLP, NLTK, customer feedback and prioritizing responses based
VEDER, RoBERta, Lexicon, Hugging face, on the expressed sentiment. By identifying
dissatisfied customers and addressing their concerns
promptly, businesses can improve their overall
I.1 INTRODUCTION customer experience and build stronger relationships.
Sentiment analysis, also known as opinion mining, is a
crucial task in natural language processing that involves Social Media Monitoring: With the rise of social
determining the sentiment, attitude, or opinion expressed in media platforms, sentiment analysis provides
a given text. This computational technique enables businesses with insights into public opinion, trends,
organizations to gain insights from vast amounts of textual and influencers. It enables them to track their brand
data and understand the sentiment of their customers or the reputation, engagement levels, and identify any
general public towards their products, services, or any other potential PR crises. By understanding how customers
subject of interest. By examining language patterns, perceive their brand and the sentiments surrounding
sentiments can be categorized into positive, negative, or it, businesses can make informed decisions regarding
neutral. Sentiment analysis finds its applications in various their social media strategies.

1
 1. VEDER (Valence Aware Dictionary for Sentiment
Market Research: Sentiment analysis is widely used Reasoning)
in market research to analyze customer sentiments
towards a particular product or service. This analysis Valence Aware Dictionary for Sentiment Reasoning
helps businesses understand market trends, identify (VEDER) is a tool used for sentiment analysis, which
involves analyzing and determining the sentiment
gaps in the market, and develop strategies to meet
expressed in a given text. VEDER is designed to
customer needs. By leveraging sentiment analysis understand the valence (positive, negative, or neutral) and
data, companies can create effective marketing arousal (degree of intensity) of words in context. By
campaigns and product development plans that align using a dictionary-based approach, VEDER assigns
with customer preferences. scores to words based on their sentiment, allowing for a
comprehensive understanding of the emotional tone of a
It's important to note that sentiment analysis can be text. This tool has proven to be effective in a wide range
performed using various methods and techniques, of applications, such as social media monitoring, brand
including natural language processing (NLP), reputation management, and customer feedback analysis.
With VEDER, businesses can gain valuable insights into
machine learning, and statistical analysis. Natural
public opinion and tailor their strategies accordingly.
language processing algorithms analyze the semantic
and syntactic structure of text to extract meaning and VEDER employs advanced algorithms to perform
sentiment. Machine learning models are trained on sentiment analysis on text. These algorithms are based on
labeled data to classify new text inputs into positive, a comprehensive dictionary that contains words and their
negative, or neutral sentiments. Statistical analysis associated valence or emotional intensity scores. These
involves measuring the frequency and distribution of scores help determine the sentiment expressed by the text.
sentiment words and phrases in a given text.
One of the key features of VEDER is its ability to
II LITERATURE SURVEY consider the context and valence shift of words. It takes
into account not only the individual sentiment scores of
2.1 RESEARCH PAPERS
words but also how their meanings change when they
appear in different contexts. This allows VEDER to
The author in [1] “Sentiment Analysis of Twitter Data” accurately capture the sentiment expressed in complex
uses SVM to analyze the text data of comments. sentences.

In paper [2] “Sentiment Analysis and Subjectivity” we The underlying algorithms of VEDER use techniques
learnt Feature-based sentiment analysis, Document-Level such as word embedding and machine learning to process
Sentiment Classification and Classification Based on and analyze the text. By leveraging large amounts of
Supervised Learning labeled data, VEDER has been trained to accurately
classify text into different sentiment categories, such as
In paper [3] “Twitter Sentiment Analysis” we found positive, negative, or neutral.
that VEDER is the pretrained model used for sentiment
analysis that uses lexicon-based tool.

In paper [4] “sentiment analysis model for text

sentiment analysis based on RoBERTa” we found that
RoBERTa is the pretrained model same as of VEDER
that uses Hugging face API for sentiment analysis that
uses transformer-based architecture.

2.2 METHODOLOGY

Here the methodologies used are VEDER(Valence

Aware Dictionary for Sentiment Reasoning )and
RoBERTa (Robustly Optimized BERT Approach) for the
sentiment analysis .In VEDER Lexicon tool is used and in
RoBERTa Transformer architecture with Hugging face
API is used.

2
1.1 Lexicon Dictionary compared to its predecessor, BERT. These optimizations
result in a more robust and powerful language model that
exhibits superior performance on a wide range of
downstream tasks
RoBERTa's adoption of the BERT architecture and its
comprehensive pre-training process makes it versatile and
adaptable to various domains and languages. Researchers
and developers can fine-tune RoBERTa for specific tasks
by training it on labeled data, allowing it to achieve state-
of-the-art performance in areas like sentiment analysis,
document classification, and question answering
The Lexicon dictionary used in VEDER (Valence Aware
Dictionary for Sentiment Reasoning) is a crucial 2.1 Fine-Tuning
component that helps in sentiment analysis. It consists of
a comprehensive collection of words and phrases
organized based on their emotional valence, or the
positive, negative, or neutral connotation they carry.

This dictionary enables VEDER to recognize and

understand the sentiment expressed in text by assigning
valence scores to each word. These scores indicate the
degree of positivity or negativity associated with the
word. By measuring the aggregated sentiment of each
word in a text, VEDER can determine the overall A fine-tuned model in Sentiment Analysis using
sentiment of the text, allowing for more accurate RoBERTa (Robustly Optimized BERT Approach) is a
sentiment reasoning and analysis. powerful tool that has been trained to accurately classify
the sentiment of text. RoBERTa, a variant of the BERT
The Lexicon dictionary in VEDER is continuously (Bidirectional Encoder Representations from
updated and refined to include new words, slang, and Transformers) model, has been optimized and fine-tuned
cultural expressions. This ensures that the sentiment to yield even better results in sentiment analysis tasks.
analysis performed by VEDER remains up-to-date and Fine-tuning involves training the model on specific
relevant in today's ever-evolving linguistic landscape. sentiment analysis datasets, enabling it to learn the
The use of this robust dictionary enhances the accuracy nuances of sentiment and make more accurate
and reliability of sentiment analysis outputs, benefiting predictions. This fine-tuned RoBERTa model takes into
various applications such as social media monitoring, account the context, semantics, and syntax of the text,
customer feedback analysis, and brand perception studies. allowing it to understand and interpret the sentiment
expressed more effectively. With its enhanced
2. RoBERTa (Robustly Optimized BERT Approach) capabilities, a fine-tuned model using RoBERTa has the
potential to greatly improve sentiment analysis
RoBERTa's architecture is based on the Transformer applications in various domains.
model, which leverages self-attention mechanisms to
process and understand text. It consists of multiple layers 2.2 Transformer Architecture
of self-attention and feed-forward neural networks,
enabling it to capture intricate patterns and dependencies The Transformer architecture has proven to be highly
within the input text. effective in the field of sentiment analysis, with one
prominent implementation being RoBERTa (Robustly
Before being fine-tuned for specific tasks, RoBERTa Optimized BERT Approach). RoBERTa is a powerful
undergoes a pre-training process to learn contextual natural language processing model that builds upon
representations of words. This process involves training BERT's success by incorporating additional training data
the model on a huge amount of unlabeled text data, such and fine-tuning techniques. By using the Transformer's
as books, articles, and web pages. By predicting missing self-attention mechanism, RoBERTa is able to capture
words in sentences, RoBERTa learns to encode the and analyze the contextual relationships between words
meaning and context of words more effectively. in a given text. This enables it to better understand the
sentiment expressed in a sentence, as it can account for
During the pre-training phase, RoBERTa follows a the dependencies and nuances present across the text.
masked language modeling objective, where random With its ability to handle large amounts of data and its
tokens in the input are masked, and the model is trained advanced training techniques, RoBERTa has emerged as
to predict the correct tokens. This task enables RoBERTa a state-of-the-art model for sentiment analysis, delivering
to grasp the relationships between words and improve its accurate and robust results.
understanding of sentence structures.

RoBERTa's pre-training process is also enhanced with

additional training data and longer training times

3
III FEATURES online reviews, businesses can gain a deeper understanding
of how consumers perceive their products or services.
1. VEDER
A key feature of VEDER is its comprehensive lexicon Identifying emerging trends
and sentiment word lists. These resources contain a In a rapidly changing market landscape, staying ahead of
vast collection of words and phrases associated with emerging trends is crucial for businesses' success. Sentiment
emotions, allowing the model to accurately detect and analysis can help organizations identify and monitor
classify emotions in visual content. The lexicon is emerging trends by analyzing customer sentiments and
continuously updated and enriched to capture conversations online. By tracking sentiment patterns
emerging emotional expressions and account for associated with specific products, services, or industry
nuances in different languages and cultural contexts. topics, businesses can gain early insights into upcoming
market trends.
The sentiment word lists in VEDER are carefully
curated and categorized, providing a fine-grained Identifying and addressing customer concerns
understanding of emotions expressed in visual content. By monitoring social media sentiment, organizations can
These lists include words with positive, negative, and identify negative sentiments expressed by customers and
neutral valences, enabling the model to distinguish promptly respond to their complaints or issues. This
between different emotional states with high accuracy. proactive approach shows customers that their concerns are
being heard and addressed, resulting in higher customer
2. RoBERTa satisfaction and loyalty. Moreover, by analyzing sentiment
RoBERTa, which stands for Robustly Optimized patterns, businesses can identify recurring issues and
BERT Approach, is a state-of-the-art language model implement long-term solutions to address them effectively.
that excels in various natural language processing
tasks. One of its key features is its language modeling
capability. RoBERTa is trained on a massive amount V Similarities between VEDER and RoBERTa
of text data to understand the structure, context, and
meaning of sentences. This enables it to generate high- The VEDER model and RoBERTa share a number of
quality representations of words and sentences, which similarities in their architecture and functionality, making
are essential for performing tasks like text
them both highly effective in natural language processing
classification, information retrieval, and sentiment
tasks. Here are some key similarities between the two:
analysis.
RoBERTa also incorporates masked language
Transformer-based Architecture: Both VEDER and
modeling, which involves randomly masking out
RoBERTa are built on the transformer architecture, which
certain words in a sentence and predicting them based
utilizes self-attention mechanisms to capture the
on the surrounding context. This helps the model learn
relationships between words in a sentence or document.
the relationships between words and improve its
understanding of the overall sentence meaning. By
combining language modeling and masked language Pre-training and Fine-tuning: Both models follow a two-
modeling, RoBERTa achieves a deeper understanding step process of pre-training and fine-tuning. Pre-training
of textual content and delivers more accurate results in involves training the models on a large corpus of text
a wide range of NLP tasks. containing general language knowledge, while fine-tuning
involves training on specific downstream tasks.

IV BENEFITS Large-scale Training Data: Both VEDER and RoBERTa

Enhancing customer satisfaction benefit from being trained on large-scale datasets, allowing
Sentiment analysis offers valuable benefits for businesses, them to learn from a wide range of linguistic patterns and
particularly in enhancing customer satisfaction. By improve their language understanding abilities.
analyzing customer feedback, social media posts, and online
reviews, organizations can gain insights into the sentiments Language-Agnostic: Both models are language-agnostic,
and emotions expressed by their customers. This meaning they can be used for a variety of natural language
understanding allows businesses to identify areas of processing tasks across different languages. This flexibility
improvement and address customer concerns more makes them suitable for multilingual applications.
effectively. By addressing these concerns promptly,
businesses can enhance customer satisfaction and loyalty.
VI Limitations and Challenges
Gaining insights into consumer behavior Despite the effectiveness and versatility of VEDER and
Market research plays a crucial role in helping organizations RoBERTa in natural language processing tasks, there are
understand consumer behavior and make informed business still some challenges faced when applying these models to
decisions. Sentiment analysis can significantly enhance sentiment analysis. Some of these challenges include:
traditional market research methodologies by providing
valuable insights into consumer sentiments and emotions. Labeling Bias: Sentiment analysis datasets are often subject
By analyzing customer feedback, social media posts, and to labeling bias, where the sentiment labels assigned to the
text might be subjective and vary across annotators. This

4
labeling bias can impact the accuracy and generalizability of nuanced language patterns and context, enabling accurate
the sentiment analysis models trained using these datasets. sentiment analysis.
Domain Adaptation: Sentiment analysis models trained on
one domain might not generalize well to other domains. VIII FUTURE SCOPE
This is because the sentiment expressions and contexts can The future scope of sentiment analysis using VADER
vary across different domains, such as movie reviews, (Valence Aware Dictionary for Sentiment Reasoning) and
product reviews, social media posts, etc. Adapting the RoBERTa (Robustly Optimized BERT Approach) looks
models to specific domains or continuously updating the promising. VADER is a rule-based sentiment analysis tool
models with domain-specific data can help improve their that uses a lexicon-based approach to analyze sentiment in
performance in sentiment analysis. text, while RoBERTa is a popular deep learning model
pretrained on a large amount of data to better understand the
Irony and Sarcasm: Sentiment analysis models often context of the text.
struggle with detecting irony and sarcasm in text, as these
expressions require a deep understanding of context and Combining the strengths of VADER's rule-based approach
subtle language cues. Irony and sarcasm detection still pose and RoBERTa's deep learning capabilities can lead to more
challenges for VEDER and RoBERTa, and improvements in accurate sentiment analysis results. The use of VADER can
contextual understanding could help mitigate these help improve the initial sentiment classification by
challenges. providing a baseline understanding of sentiment based on
predefined rules and word valence scores. This can be
Out-of-Vocabulary Words: Sentiment analysis models like particularly useful for domains or languages where large
VEDER and RoBERTa heavily rely on pre-training on vast amounts of labeled data for training deep learning models
amounts of text data. However, they may still struggle with might be scarce.
out-of-vocabulary words or uncommon expressions that
were not encountered during training. These models might On the other hand, RoBERTa can enhance sentiment
not accurately capture the sentiment of such words and analysis by capturing the contextual nuances and
expressions, leading to potential errors in sentiment complexities of text. Its ability to comprehend the broader
analysis. meaning of words and phrases in their specific context can
lead to a more nuanced and accurate sentiment
Contextual Understanding: Sentiment analysis tasks often classification. By leveraging RoBERTa's pretraining and
require a strong understanding of the context in which the fine-tuning capabilities, sentiment analysis models can be
sentiment is expressed. VEDER and RoBERTa excel in trained to better understand the subtleties of sentiment
capturing contextual relationships, but they may still face expressed in text across different domains and languages.
challenges in accurately interpreting complex contexts or
subtle nuances present in certain texts. In summary, the future of sentiment analysis using VADER
and RoBERTa holds great potential in achieving more
VII CONCLUSION accurate and robust sentiment analysis results. By
combining the strengths of a rule-based approach and deep
In conclusion, VEDER and RoBERTa are highly effective learning, these models can provide a more nuanced
and versatile models for sentiment analysis in natural understanding of sentiment in text, leading to improved
language processing tasks. VEDER leverages deep sentiment analysis applications in various domains and
learning techniques and visual features from images to languages.
extract sentiment information from multimodal data, while
RoBERTa utilizes transformer-based language models for REFERENCES
robust language understanding. These models have
demonstrated impressive performance in various sentiment [1] Sentiment Analysis of Twitter Data by Apoorv Agarwal
analysis tasks and have been widely adopted in both Boyi Xie Ilia Vovsha Owen Rambow Rebecca Passonneau
academia and industry. Department of Computer Science, Columbia University
New York, NY 10027 USA
VEDER combines visual and textual information through
the fusion of pre-trained deep visual features and attention [2] Sentiment Analysis from User-Generated Reviews of
mechanisms. It has shown promising results in capturing Ride-Sharing Mobile Applications
visual sentiment information from images, making it a https://ieeexplore.ieee.org/abstract/document/9753947
reliable model for sentiment analysis in multimedia
content analysis. [3] FinTech: Deep Learning-Based Sentiment
Classification
https://link.springer.com/chapter/10.1007/978-3-031-
RoBERTa, on the other hand, excels in language
understanding and context modeling. Its pre-training 38296-3_10
process with large-scale unlabeled data helps it capture