Text Sentiment Analysis

Submitted By: Mentored By: Dr. Jay Shankar Prasad

● Gaurav - 1901320100062
● Bhavya Tyagi - 1901320130029 Associate Professor
● Manish Kumar Gupta - 1901320100087 Greater Noida Institute of Technology
● Riya - 1901320100128
Problem Statement
● Analyzing sentiments of people plays an important role in understanding customer opinions and feedbacks to
improve products, services, and customer experience.
● Brand Reputation Management by monitoring and analyzing online conversations and social media posts which can
help the companies to timely address all negative reviews about them, Marketing and Advertising, Crisis
Management, and even in Political and Social Analysis which can be helpful in reducing negative effects of Social
unrest and provide valuable insights for policymakers and experts.
● But, these models are made to work for a specific domain, for example, Market Specific, Movie or other sentiment
Analysis models, so there is a need for Generalized Sentiment Analysis models which is trained on data from
different domains.
● Therefore, to address this issue, we need a system that is able to analyze data from different domains.
● So, in order to create such system, we have used training data from different domains. Dataset constitute of data from Social
media site (twitter in this system), and Airline reviews.
● This system is helpful in analyzing sentiments of dataset, of paragraphs too which user can type on their own or copy from any
source.
Abstract
● Sentiment analysis is the process of using natural language processing, text analysis, and statistics to
analyze customer sentiment.
● Sentiment analysis (SA) is a process of extensive exploration of data stored on the Web to identify and
categorize the views expressed in a part of the text. It is the study of automated techniques for
extracting sentiments from written languages.
● The intended outcome of this process is to assess the author attitude toward a particular topic, movie,
product, etc. The result is positive, negative, or neutral.
● With increasing popularity of internet users has been growing fast parallel to emerging technologies
that actively use online review sites, social networks and personal blogs to express their opinions the
need of Sentiment analysis has also been increasing.
● These data and information can potentially be utilized to provide real-time insights into the sentiments
of people which can be helpful in Product analysis, Customer service Response, Social media
monitoring and brand management.
● It is extensively used in fields like data mining, web mining, and social media analytics.
Sentiment Analysis reviews used to understand what people at large are trying to communicate to the
businesses and therefore its analysis becomes important.
Main Objectives
● The main objective of Sentiment analysis is to accurately extract people's opinions from a large
number of unstructured review texts and classifying them into sentiment classes, i.e., positive,
negative, or neutral.
● Sentiment analysis is also called opinion mining, which refers to the use of
natural language processing and text mining to identify the emotional information from text materials.
● Opinion Mining can be helpful in :
○ It is a subjective assessment of something based on personal empirical experience. It is partially
rooted in objective facts and partly ruled by emotions.
○ An opinion can be interpreted as a sort of dimension in the data regarding a particular subject. It
is a set of signifiers that in combination present a point of view, i.e., aspect for the particular
issue.
○ It can also find and extract the opinionated data on a specific platform, determine its polarity
(positive or negative), define the subject matter (which can help in research work too) and
identify the opinion holder (on its own and in correlation with the existing audience segments)
● Sentiment analysis is the procedure by which information is extracted from the opinions, appraisals
and emotions of people in regards to entities, events and their attributes.
Introduction

● Sentiment analysis (or opinion mining) Sentiment analysis or opinion mining is the computational
study of people’s opinions, sentiments, attitudes, and emotions expressed in written language. It is a
natural language processing (NLP) technique used to determine whether data is positive, negative or
neutral.
● Sentiment classification is a way to analyze the subjective information in the text and then mine the
opinion. Sentiment analysis is the procedure by which information is extracted from the opinions,
appraisals and emotions of people in regards to entities, events and their attributes.
● It is often performed on textual data to help businesses monitor brand and product sentiment in
customer feedback, and understand customer needs.
● It is used by business and data analyst to detect sentiment in social data, gauge brand reputation, and
understand customers.
● It helps in, first, decision making based on customer or end user opinions have a significant effect on
customers ease, and second, it presents many challenging research problems by studying customer’s
opinions and suggestions.
Methodology
● Data Cleaning: This process involves pre-processing of data, i.e., removal of stop words like as, and, or, as these words do
not contribute much in analysis and removing them reducing data size.
● Categorical Encoding: This involves converting categorical data into numerical one, as machines can only understand
numbers.
● Preparing Data for Model: All sentences in the dataset are converted to a sequence of integers. All sequences have same
length, which is equal to the length of longest sequence in the training dataset.
● Neural Architecture: LSTM (Long Short Term Memory) is used for the processing of the data, here LSTM layer has 32
memory cells or hidden units.
○ Output of Embedding layer is taken as input by each layer.
○ Embedding layer just learns the vector or integer representation for each word. After this mapping of vector of arbitrary
real valued numbers to probability distribution.
● After training phase, model weights are fixed (they are fixed or set as the one which gives highest accuracy for input data). The
model then uses these learned weights to process test data and provide results, which are to predict whether sentence is
positive, negative or neutral.
Flow Chart
Activity Diagram
DATA Flow Diagram
Literature Review
● Universal Facial Expression of Emotions [1]: In 1970, psychologist Paul Ekman and his colleagues Wallace Friesen
published their study on the Universal Expression of emotions. It was based on the
○ Hypothesis: There are six universally recognized emotions: surprise, anger, fear, disgust, sadness, and happiness.
○ Experiment: 30 photographs displaying facial expressions of these six emotions were shown to people living in Brazil,
The United States, Argentina, Chile and Japan to predict the emotions or sentiments associated with the expression.
○ Results: The study proved that certain facial expressions are universally recognized.
● Human Speech Emotion Recognition [2]: Maheshwari Selvaraj research focused on developing efficient and accurate
models to recognize emotions from speech signals.
○ This research involves finding acoustic features which can be used in extracting sentiments of the speaker.
○ This research utilizes techniques like feature extraction, signal processing for analyses of speech signals and recognize
emotions.
○ This research address key challenges associated with speech based sentiment analysis, which are missing data due to
signal loss, inter and intra speaker variability.
● Audio Visual Modalities for Emotion Recognition by De Silva and others [3]: This research describes a multi-modal
approach that combines facial expressions and speech signals, for emotion detection.
○ Feature Extraction: features from both the speech signals and facial expression of the participants while they were
watching emotional video clips are extracted
○ Results: Features extracted were used to train Emotion recognition models. Results showed Bimodal systems results in
higher accuracy than any Unimodal systems.
Literature Review
● Audio Visual Modalities for Emotion Recognition by Chen and others [4]: This research describes a multi-modal sentiment
analysis by using Electroencephalogram (EEG) signals in addition with audio and visual modalities.
○ Feature Extraction: Features from EEG, speech, and facial signals, that were recorded while participants watched an
emotional video clip were recorded.
○ Results: Features extracted were used to train Emotion recognition models. Results showed Bimodal systems results in
higher accuracy than any Unimodal systems.
● Sentiment Analysis in English Texts Arwa Alshamsi [5]: This research involved comparing performances of various Machine
Learning Algorithms for Analysis of Sentiments, like, Support Vector Machines (SVMs), Random Forest, Naïve Bayes, and
Decision Trees.
○ It also evaluated impact of feature selection techniques, like, term-frequency-inverse document frequency (TF-IDF).
Results of the study showed SVM algorithm performed best on the data set consisting of hotel reviews.
○ Results: This research proved the importance of selecting right machine learning algorithm, feature selection technique.
● Sentiment Analysis using Product Review Data by Xing Fang & Justin Zhan [6]: This research involved comparing
performances of various Machine Learning Algorithms for Analysis of Sentiments, like, Support Vector Machines (SVMs),
Random Forest, Naïve Bayes, and Decision Trees.
○ It also evaluated impact of feature selection techniques, like, term-frequency-inverse document frequency (TF-IDF).
Results of the study showed SVM algorithm performed best on the data set consisting of hotel reviews.
○ Results: This research proved the importance of selecting right machine learning algorithm, feature selection technique
Literature Review
● Missing Modality Robust Emotion Recognition Framework by Sun and others [7]: This research aimed at resolving the
problem occur due to missing modalities in emotion recognition, which is due to unavailability of sufficient data.
○ Approach Used: Iterative data augmentation to generate synthetic data that fill the gaps in missing modalities. Repetition
of this process occur iteratively until satisfactory level of performance is gained.
○ Results: This approach is mainly focuses on addressing missing modality problem in sentiment analysis.
● Aspect Based Sentiment Analysis by Tao Yang, Qing Yin, Lei Yang, and Ou Wu [8]: This study aims at improving aspect-
based analysis of sentiment using new target representation and dependency attention.
○ The dependency attention mechanism uses graphical neural networks to model dependency between words in text.
These graph neural networks contain information from both word and target embedding.
○ Results: This study showed the importance of incorporation of target representation and dependency attention mechanism
in aspect based analysis of sentiments
Screenshots

Fig.User Interface
Screenshots

Fig.Output Interface for File page

Screenshots

Fig.Output Interface for Text page

Future Scope
● It can be used for monitoring social media apps at real time.
● Analysing not only the sentiment but along with that suggestions too to solve those problems, for
example, if most of the reviews for a mobile company are about its lower battery backup, then it can
analyse the internet and provide solution to company by suggesting them how to solve this issue with
resources in hand.
● For Understanding and predicting political opinions and polls, keeping a check on social unrest,
Preventing PR crises, increasing morale and boosting productivity by listening to the employees,
predict and analyses market trends
● Examine social media conversations about your brand using sentiment analysis to see what your
customers are saying.
● Analyse large amounts of employee feedback data to determine employee satisfaction levels.
● Analyse sarcastic comments or comments made in different languages, along with emoji used.
Conclusion
In this Project, we have used labelled data to design a model for analysis of text sentiments (by Sentiment
Analysis we meant the process of drawing out subjective information from written language which can
further be used in business or other analysis to understand people’s opinions, sentiments, attitudes, and
emotions), which can predict the sentiments of the text and label them into multi-class sentiments, which are
‘Positive’, ‘Negative’, and ‘Neutral’. For this LSTM is used for training the model.
References
1. Paul Ekman and others: Universal facial expressions of emotion, Culture and Personality: Contemporary Readings/Chicago,
pp. 151–158, 1974.
2. S. Maheshwari , R. Bhuvana, S. Padmaja, Human Speech Emotion Recognition, / International Journal of Engineering and
Technology (IJET), Vol 8 No 1, pp.311-323, p-ISSN : 2319-8613
3. L. C. De Silva and Pei Chi Ng, "Bimodal emotion recognition," Proceedings Fourth IEEE International Conference on Automatic
Face and Gesture Recognition (Cat. No. PR00580), Grenoble, France, 2000, pp. 332-335, doi: 10.1109/AFGR.2000.840655.
4. L. S. Chen, T. S. Huang, T. Miyasato and R. Nakatsu, "Multimodal human emotion/expression recognition," Proceedings Third
IEEE International Conference on Automatic Face and Gesture Recognition, Nara, Japan, 1998, pp. 366-371, doi:
10.1109/AFGR.1998.670976.
5. A. Alshamsi, R. Bayari, S. Salloum "Sentiment Analysis in English Texts", Advances in Science, Technology and Engineering
Systems Journal, vol. 5, no. 6, pp. 1683-1689 (2020).
6. X. Fang, J. Zhan, Sentiment analysis using product review data. Journal of Big Data 2, 5 (2015).
https://doi.org/10.1186/s40537-015-0015-2
7. Wang, Ning. “M2R2: Missing-Modality Robust emotion Recognition framework with iterative data augmentation.” ArXiv
abs/2205.02524 (2022): n
8. T. Yang, Q. Yin, L. Yang , O. Wu, Aspect-Based Sentiment Analysis with New Target Representation and Dependency
Attention,IEEE Transactions on Affective Computing, vol. 13, no. 2, pp. 640-650, 1 April-June 2022, doi:
10.1109/TAFFC.2019.2945028.
References
9. M. Wollmer et al., “Youtube movie reviews: Sentiment analysis in an audio-visual context,”IEEE
Intell. Syst., vol. 28, no. 3, pp. 46–53, May/Jun. 2013.

10. A. Metallinou, S. Lee, and S. Narayanan, “Audio-visual emotion recognition using Gaussian
mixture models for face and voice,” in Proc. 10th IEEE Int. Symp. ISM, 2008, pp. 250–257.

11.https://deepai.org/publication/multimodal-sentiment-analysis-addressing-key-issues-and-setting-up-
baselines

12.https://www.researchgate.net/publication/360410424_M2R2_Missing-
Modality_Robust_emotion_Recognition_framework_with_iterative_data_augmentation

13.https://blog-knoldus-com.cdn.ampproject.org/v/s/blog.knoldus.com/what-are-transformers-in-nlp-
and-its-advantages/amp/?amp_gsa=1&amp_js_v=a9&usqp=mq331AQKKAFQArABIIACAw%3D
%3D#amp_tf=From%20%251%24s&aoh=16706823819170&referrer=https%3A%2F
%2Fwww.google.com&ampshare=https%3A%2F%2Fround-lake.dustinice.workers.dev%3A443%2Fhttps%2Fblog.knoldus.com%2Fwhat-are-transformers-
in-nlp-and-its-advantages%2F