0% found this document useful (0 votes)
344 views17 pages

Afaan Oromo Question Classification Using Deep Learning Approachosal

This proposal outlines a system to detect negative emotions in Amharic texts on social media using sentiment analysis and deep learning. As social media usage grows in Ethiopia, it has also fueled divisions between people. The government has responded by restricting access. The proposed system would analyze Amharic language posts to identify negative emotions, which could help address issues before escalation. It would apply techniques like sentiment analysis and deep learning models since Amharic is under-resourced for natural language processing. The system aims to contribute to more peaceful social media discourse in Ethiopia.

Uploaded by

olana temesgen
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
344 views17 pages

Afaan Oromo Question Classification Using Deep Learning Approachosal

This proposal outlines a system to detect negative emotions in Amharic texts on social media using sentiment analysis and deep learning. As social media usage grows in Ethiopia, it has also fueled divisions between people. The government has responded by restricting access. The proposed system would analyze Amharic language posts to identify negative emotions, which could help address issues before escalation. It would apply techniques like sentiment analysis and deep learning models since Amharic is under-resourced for natural language processing. The system aims to contribute to more peaceful social media discourse in Ethiopia.

Uploaded by

olana temesgen
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 17

CPU Business and Information Technology College

School of Postgraduate Studies

Department of Software Engineering

NEGATIVE EMOTION DETECTION IN SOCIAL MEDIA FOR

AMHARIC TEXT USING SENTIMENT ANALYSIS AND DEEP

LEARNING METHODS

A Proposal Submitted to Department of Software Engineering in Partial Fulfillment of the


Requirements for the Degree of Masters of Science in Software Engineering

By

Abel Tibebu

Advisor: Sileshi Yalew (PhD)

Addis Ababa, Ethiopia

Date: October, 2022


Table of Contents
List of Acronyms and Abbreviations ............................................................................................................ 3
Abstract ......................................................................................................................................................... 4
CHAPTER 1: INTRODUCTION ................................................................................................................. 5
1.1 Background of the Study .................................................................................................................... 5
1.2 Statement of the Problem .................................................................................................................... 6
1.3 Significance of the Study .................................................................................................................... 7
1.4 Research Questions ............................................................................................................................. 7
1.5 Objectives of the study........................................................................................................................ 8
1.6 Scope and Delimitations of the Study ................................................................................................. 8
1.7 Research Methodology ....................................................................................................................... 9
CHAPTER 2: LITERATURE REVIEW .................................................................................................... 10
2.1 Related Works ................................................................................................................................... 10
2.1.1 Sentiment analysis in local language ......................................................................................... 10
2.1.2 Emotion classification in the English language ......................................................................... 12
CHAPTER 3: PROPOSED METHODS..................................................................................................... 14
3.1 Data Collection and Preparation ....................................................................................................... 14
3.2 Techniques ........................................................................................................................................ 14
3.3 Tools ................................................................................................................................................. 15
3.4 Evaluation Technique ....................................................................................................................... 15
CHAPTER 4: WORK PLAN...................................................................................................................... 16
4.1 Research Schedule ............................................................................................................................ 16
4.2 Research Budget ............................................................................................................................... 16
References ................................................................................................................................................... 17

2
List of Acronyms and Abbreviations

AI – Artificial Intelligence

ANN – Artificial neural network

BI-LSTM – Bidirectional long-short term memory

CNN – Convolutional neural network

DNN – Deep neural network

LSTM – Long-short term memory

ML – Machine learning

MLP – Multi-Layer Neural Networks

NLP – Natural Language Processing

NLTK – Natural Language Toolkit

NR – Neural Network

NB – Naïve Bayes

RNN – Recurrent Neural Networks

SVM – Support Vector Machine

TF-IDF – Term Frequency- Inverse Document Frequency

3
Abstract
In the last four years, the ever growing usage of social media in Ethiopia has fueled the country‘s
problem against the peaceful coexistence of its people. Illegitimate social media usage has played
a significant role in widening the distress between the people. As a result, the government has
increasingly relied on the temporary closure of social media sites; nationwide internet shutdowns
and filtering websites to suppress polarizing voices. Many studies have shown that social media is
an arena for sharing information that reflects negative emotions. People express emotions as part
of everyday communication. Emotions can be judged by a combination of cues such as facial
expressions, prosodies, gestures, and actions. Emotions are also articulated by written texts as most
people would like to post their articles in social network to express their feeling, it would benefit
to collect and analyze these information to figure some signs before some misfortunes happened.
However, low resource languages such as Amharic have received less attention due to several
reasons such a slack of well-annotated datasets, unavailability of computing resources, and only
few researches are done in the area. In this paper, I propose a system that detect negative emotions
in Amharic texts using sentiment analysis and deep learning approaches.

Keywords: - Emotional analysis, Sentiment analysis, Machine learning, Deep learning

4
CHAPTER 1: INTRODUCTION
1.1 Background of the Study

Social media is computer-based technology that facilitates the sharing of ideas, thoughts, and
information through the building of virtual networks and communities. By design, social media is
internet-based and gives users quick electronic communication of content. Content includes
personal information, documents, videos, and photos. Users engage with social media via
computer, tablet or smartphone via web-based software or web application, often utilizing it for
messaging. Social media typically features user-generated content and personalized profiles.
Content can be information, entertainment or nothing specific at all. Content comes in various
forms: video, audio, text or image. It will be designed to transfer a certain feeling, information and
data to somebody. Furthermore, the overuse of social media is a global problem impacting all
generations, and research has shown that substantial Internet usage can have a highly negative
impact on our mental and emotional health. Symptoms of anxiety, depression and obsessive-
compulsive disorder can be triggered by the overuse of social media, as individuals are constantly
concerned about their posts and communicating with others. When it comes to the experience of
the society as a whole, social media has enormous negative impacts. Some of the disadvantages to
society as indicated in the stat-counter report are cyberbullying, Hacking, Addiction, Fraud, and
others [1].

The content on social media that is composed, transmitted, accessed, or received may contain
contents that could be considered discriminatory, offensive, obscene, threatening, harassing,
intimidating, or disruptive to any person. Examples of unacceptable content may include, but are
not limited to, sexual comments or images, racial slurs, gender -specific comments, or any other
comments or images that could reasonably offend someone on the basis of race, age, gender,
religious or political beliefs, national origin, disability, sexual orientation, or any other
characteristic protected by law. Because of this and many other problems, social media content
has to be filtered.

Sentiment analysis, also called opinion mining, will be the field of study that analyzes people‘s
opinions, sentiments, evaluations, appraisals, attitudes, and emotions towards entities such as
products, services, organizations, individuals, issues, events, topics, and their attributes[2].

5
On social media, the sentiment of a post can be seen in the tone or emotion conveyed in a text. We
can understand exactly how people feel about political, social, religious situations in Ethiopia, by
effortlessly analyzing the sentiment of each and every post and monitoring how the community is
responding to contents. Social media sentiment analysis involves applying natural language
processing (NLP) to social mentions from various sources and determining whether the user is
talking about political situation, religious organization or some products in a positive, negative or
neutral way. A proper social media sentiment analysis could categorize social media mentions into
the right category. Social media sentiment analysis uses some revolutionary machine learning and
deep learning algorithms and analyses the text posted online. Sentiment analysis is a challenging
task especially for languages having low resources and complex linguistic structures like Amharic.
A small amount of resources like training data for Sentiment analysis highly impacts the accuracy
of the system. The complex structure of languages also needs the design of important features and
the best combination from these features. Amharic as one of low resourced and morphologically
rich languages shares the above challenges. The problem is information filtering and sentiment
analysis problem, and our goal is to investigate which deep learning methods are best suited to
solve it.

1.2 Statement of the Problem

Social media authors can upload different content to the internet, and viewers can react to any type
of attitude, abuse, and unrelated text in the content by commenting on it. Due to the rapid
development of emotional comments, reports, and posts on the Internet, there is a need for
obtaining important sources, finding related posts or comments with negative emotion, analyzing,
and organizing them into the appropriate form. Furthermore, researches done on Sentiment
analysis (SA) up to now are specific to high resourced languages like English, French and other
European languages but as Amharic is a low-resource language, in regards to digitization, there is
little attempt on this area. Therefore, in this study the performance of sentiment analysis and deep
learning algorithms on Amharic language will be analyzed and explored. Based on sentiment
analysis of texts on social media, negative emotion detection solution can protect against Toxic
online content or can be used to tackle politically and socially sensitive contents, and prevent
illegal or unsuitable social media content from being accessed.

6
1.3 Significance of the Study

In the completion of this research on “Negative Emotion Detection in Social Media for Amharic
Text Using Sentiment analysis and Deep learning Methods” will have some significance to the
development of online sharing activities and also give some benefit to society, specifically for
social media authors, commenters, psychologists, and government. One very practical use in my
study area, People pay attention to comments and posts and buy more of the attention that has the
most participation going on and using sentiment analysis, we will be alerted to negative sentiments
as they happen, allowing you to respond quickly.

The Negative Emotion Detection model can be used for different purposes. Some of them are:

• Government organizations can use the system to reduce political, social and religious
turbulences.

• The system can be used to answer sentiment questions. For instance, what is the social
media user’s reaction to the speech by the prime minister?

• It has a huge role for psychologists who want to understand specific emotion from
written text

• Companies can use sentiment analysis to gain deep insight into what‘s happening
across their customer support.

1.4 Research Questions

The study will attempt to explore and answer the following research questions.

RQ1. How to identify and extract negative emotions from social media content?

RQ2. How to annotate social media content in to negative emotion and positive emotion content?
RQ3. To what extent can deep learning approach improve the performance of the negative
emotion detection from Amharic texts?

RQ4. How to select the best deep learning algorithm for detecting negative emotion content?

RQ5. What are the main challenges in Amharic sentiment analysis?

7
1.5 Objectives of the study

General Objective

The main goal of this research is to detect negative emotion contents (text) in social media for
Amharic texts based on sentiment analysis and deep learning approaches.

Specific Objectives

 To study relevant literature in the area and draw from emotion theories in the fields of
psychology and linguistic

 To gather textual data (corpus) that contain negative emotion content

 To train a model that can detect negative emotion content from social media comment
and posts

 To validate, test and evaluate the model with respect to meeting objectives and
performance measures

1.6 Scope and Delimitations of the Study

The scope of this study is to develop a negative emotion detection model for Amharic text in social
media using sentiment analysis deep learning methods. In social media, people generally use
informal or semi-formal language for communication. In everyday life conversation, people do
not care about the spellings and accurate grammatical construction of a sentence which makes
sentiment classification a complex task. Because of this sentiment classification requires effective
analysis and processing of documents. Since there are no publicly available Natural Language
Processing (NLP) tools and other resources for Amharic sentiment classification in social media,
this study focuses on solving the problem mentioned in the problem statement that occurs in
current Ethiopian social media posts and comments related to political news or posts.

The following are considered as the limitation of this study:-

 Limited to purely Amharic texts on social media.

 Limited to sentiment classification (Positive, neutral and Negative).

8
 Limited on Negative emotion state category.

 Offensive content publisher’s identification and reasons for offensive and non-offensive
classifications are not covered in this research work.

 Limited only on two social media sites (i.e. Facebook and YouTube only).

1.7 Research Methodology

In order to accomplish the objective of this research, I will use design science research
methodology, which offers specific guidelines for evaluation and iteration within research. It
focuses on the development and performance of designed with the explicit intention of improving
the functional performance of the design.in this model all design begin with awareness of a
problem. This study will first conduct a comprehensive review of literatures to understand the
research area and its problem domains.

Based on this understanding, the study will identify the importance of the works done in the area
and analysis of the different technique, approaches and algorithms suitable and applicable for
negative emotion detection. This designation emphasizes the problem-solving or performance
improvement of the model then suggestions for a problem solutions are drawn from related works
done on this research area by assessing to identify and point direction in order to provide solution
to identified problems or improve the performance from the existing knowledge base for the
problem area.

Finally, based on the literature survey I will select appropriate tools, techniques and algorithms for
the problem. After identifying those requirements I will design the model and identify its
components for an attempt in implementing an artifact according to the suggested solution is
performed. Next preparing data properly, it’s an important and critical step in the research for data
analysis and it has an immense impact on the success of the research. I will use systematic sample
techniques. Therefore, I need to collect data from different official Facebook pages and YouTube
pages for partially or fully successful implementations and evaluation of the system using some
performance evaluation metrics like precision, F1 score, recall and accuracy. The overall
methodology is to produce artifacts including model, prototype and algorithm we also evaluate the
performance of the model through prototype.

9
CHAPTER 2: LITERATURE REVIEW

In literature review chapter, various studies of sentiment analysis both on local language and other
language using different techniques are discussed. Also, the current state of the art on sentiment
analysis and clearly define what are the gap of Amharic sentiment analysis we done extensive
review on different thesis.

There is a growing number of research works on social media emotion detection, development and
refining the automated techniques of sentiment classification and analysis. Many researchers have
worked on sentiment analysis techniques via different approaches (Lexical, Machine Learning,
and Hybrid) however, in-depth analysis and review of the latest literature on sentiment analysis
with deep learning algorithms for under-resourced language like Amharic is still required.

2.1 Related Works

In emotion classification using text, research was done in different languages, such as English,
Chinese, Arabic, and French, using different techniques and approaches. Different researches
apply different approaches, such as machine learning, ontology-based approaches, rule-based
approaches, lexicon-based approaches, feature-selection, cross validation, and others (Karna et al.,
2020). In addition to the strategies, the utilized approaches, goals, motivation, domain, target
language, dataset source, methods, accuracy, and challenges are the most focused when going to
review the various works. According to the knowledge, negative emotion detection for Amharic
texts has never been studied even though the number of negative emotional texts on the web is
increasing. In this section, I have focused on sharing and adopting the knowledge of how emotion
classification and related areas are done from resource-rich texts for different languages.

2.1.1 Sentiment analysis in local language

Wondwossen Philemon and Wondwossen Mulugeta (2015) proposed a multi-scale approach to


sentiment analysis provides a more refined breakdown than the traditional positive-negative binary
scheme. The approach is preferred in cases where comparison and ranking of opinions is vital.
They employed Naïve Bayes machine learning algorithm and used unigram, bigram and hybrid
variants as features. There Naïve Bayes implementation to multi-scale sentiment analysis was
successful and they achieved a promising performance accuracy of 43.6%, 44.3% and 39.5% for

10
unigram, bigram and hybrid language models, respectively despite the few training data used. To
achieve the objective, the author has prepared a sample corpus that contains 608 posts. The corpus
was collected from social media sources such as Facebook, Twitter, Dire Tube and Ethiopian
reporter websites [3].

According to Mihret & Atinaf, (2019), proposed a system for sentiment analysis model for
opinionated awngi text and extract the sentiment of a sentence using an tf-idf feature extraction
methodology. They analyze sentiments as positive, neutral and negative from texts based on the
opinion words representing each story sentence. In this paper, they presented a machine learning
approach to sentiment analysis on Awngi language. They used the Ethiopic script as it is by
applying the conversion to the Latin transliterated texts. In addition to this, they can achieve 75%
accuracy of the experiment is low. Information Gain feature selection method used to calculate
most informative words from the document and three supervised classifiers implemented from the
Natural Language Toolkit (the Naïve Bayes, and Maximum Entropy classifiers). The study results
obtained were encouraging. Indeed further work are initiated and forwarded [4].

Hiwot Wonago (2020) proposes Information filtering of social media Amharic texts based on
Sentiment Analysis. In this work the concept of sentiment analysis on Amharic text on social media
and presents a comparative study on machine learning algorithms. They created social media
content filtering system has been tested on Facebook posts of each class, and it has been observed
that SVM with word2vec has performed best in comparison to other classifiers, achieving average
precision of (72%), but did worse on recall (63.4%). The contents classified into “offensive” and
“non-offensive” categories. However she used small data sets which was focused only on
Facebook data collected posts/comments for the task of sentiment analysis [5].

Another study by Megersa Oljira, (2020) sentiment analysis for classifying Afaan Oromoo
Facebook Media in to positive, negative by using deep learning approach in Facebook, Twitter,
and Blogger domains. The author employed two deep learning classification techniques (Long
Short Term Memory and Convolutional Neural Network) using they used word embedding as a
feature. The system can accommodate long comments, which were difficult to categorize since it
was normal to find a contradiction in the opinion expressed, and longer phrases are based on the
context of their counterparts. The experiments are conducted using 1452 comments collected from
the official site of the Facebook page of Oromo Democratic Party/ODP for the study. Based on

11
their relative performance of classification, Convolutional Neural Network achieves the accuracy
of 89%. The Long Short Memory achieves accuracy of 87.6% .as shown from the result obtained
CNN performed better than LSTM algorithm [6].

Eyob tesfu (2021) proposed a deep learning based emotion detection for Amharic text using word
embedding and CNN. Evaluation of the system was done by comparing the two most known
classifiers CNN and RNN also with selected related works. The experimental result in this study
showed a considerable performance on CNN with Word2vec classification. The proposed model
achieved average of accuracy 71.01%, precision 71.06%, recall 71.05% and F1 score 70.96%.
However, in this study he included only four emotional states (Anger, Sadness, Happiness and
Disgust) but there are many types [7].

2.1.2 Emotion classification in the English language

The authors Tocoglu et al., (2019) proposed emotion classification for Turkey text based on deep
learning architectures, namely artificial neural network (ANN), CNN, and LSTM. The dataset
includes more than 195K documents in six emotion categories of fear, happiness, disgust, anger,
sadness, and surprise. They achieved the highest classification performance by using the CNN
architecture, which was 74% in terms of accuracy. However, they do not consider points to extend
their study in several dimensions like, to enlarge the size of the automatically constructed dataset
and word embedding schemes such as word2Vec, fastText, etc [8].

According to Park et al., (2020) they have developed a system for emotion recognition from text
stories and extract the emotion of a sentence using an embedding model. They analyze emotion
classes based on Plutchik’s wheel of emotion model from story texts based on the emotional words
representing each story sentence. The emotion in the tweet text is detected based on three elements,
namely location, event, and character. They conducted their experiments by applying conventional
neural network techniques and achieved the lowest accuracy of 36% and the highest accuracy of
73% by using 144,701 tweets, where each tweet was labeled with an emotion keyword. The
drawback of this research work was that it could not handle expressions that negate the sentences
and could not be enhanced to detect contextual emotional information in the story text [9].

Karna et al., (2020) states the Long Short-Term Memory mechanism (LSTM) deep learning
approach is based on text emotion recognition for a chatbot in English text and categorizes the text

12
into seven groups, namely Rage, Worry, Happiness, Affection, Sadness, Astonishment, and
Kindness. They experimented and found text emotion classification provides relatively higher
accuracy compared to the existing learning methods, and they have the best average performance
in terms of efficiency, sensitivity, and f1 score at 94.7%, 94.2%, and 94.1% respectively. However,
the drawback to their system was the small amount of data used as well as the fact that the data
was not annotated by experts, and nothing was mentioned about the clarity of the dataset [10].

According to Wang & Zhao, (2020) they can see from the experimental data that the BiGRU model
outperforms the Bi-LSTM model in terms of training speed. They have crawled over 4,000 pieces
of Chinese microblog data on the microblogging platform to create a microblog corpus with a high
number of emoticons. The average training speed of the Bi-GRU model is quicker than the Bi-
LSTM model when varied word vector dimensions are taken into account. This is because GRU
combines the input and forgetting gates of the LSTM into the update gate, simplifying the model
construction and increasing training speed. However, they only compare two types of deep
learning, but they should compare more than two deep learning techniques to arrive at the optimal
result. The findings from the experiment show that the BiGRU model works better in the emotional
interpretation of the Chinese text. F1 is raised by 74% compared to the BiLSTM model in
comparison to the CNN model [11].

13
CHAPTER 3: PROPOSED METHODS

A methodology is a collection of procedures, techniques, tools, and documentation aids that


support the researcher in undertaking and implementing research work [8]. This research is
experimental research, which will be designed to develop negative emotion detection from Social
Media in Amharic text. For the successful completion of this study, I will consider the following
methods.

3.1 Data Collection and Preparation

All the necessary data which will be needed to conduct this research work are post and comments
collects from Amharic social media such as various Facebook pages and YouTube channel because
there is no well-defined corpus available to classify each comments into the proper classes. The
data will be collected manually from the topics commented to social media channels. In addition,
preprocessing activities will be done on the collected data to enable language identification, data
cleaning to remove unwanted script characters, numbers, or punctuation marks from the
comments, and also removal of stop words, word tokenization, and morphological analysis.

3.2 Techniques

In this study, I will use deep learning approaches. Deep learning is a powerful set of techniques
for learning in deep neural networks. It is a branch of machine learning based on a set of algorithms
that attempt to model high-level abstractions in data by using multiple processing layers with
complex structures[9]. A deep learning algorithm automatically learns many features from big data
rather than adapting handcrafted features and is easily classified based on their category [10]. Deep
learning is a form of machine learning that needs large amounts of data to get better at functioning.
Deep learning's popularity grew as the amount of information generated expanded, as did the
advent of hardware that allowed for powerful computers. Machine learning techniques have been
outperformed by current deep learning methods. It can be applied to a wide range of problems,
such as natural language processing, fraud detection, detecting developmental delays in children,
automatic machine translation, text to image translation, market-rate forecasting, news
aggregation, and fraud news detection. There are many types of deep learning approaches, namely
recurrent neural networks, convolutional neural networks, long short-term memory, etc. That take

14
into account to preprocess of different features of raw data. For text emotion classification tasks,
various types of deep learning algorithms are applied.

3.3 Tools

I will use different tools for the accomplishment of this research.

Python: Python is a powerful scripting language and will be very useful for solving statistical
problems involving machine learning algorithms. Thus, the Python programming language will be
used for accessing important libraries and modules for development. It has various utility functions
which help in preprocessing. Processing is fast and it will be supported on almost all platforms.
Integration with C++ and other image libraries is very easy, and it has in-built functions and
libraries to store and manipulate data of all types. It provides the pandas and numpy framework
which helps in manipulation of data as per our need. A good feature set can be created using the
numpy arrays which can have n-dimensional data.
Jupyter Notebook: Jupyter notebook as a code editor; a powerful scientific environment for
Python, to develop the emotional analysis and detection from social media in Amharic text. Jupyter
Notebook is the IDE to combine python with all the libraries which will be used in implementation.
It is interactive, although some complex computations require time to complete. Plots and images
are displayed instantly. It can be used as a one stop for all our requirements, and most of the
libraries like Dlib, OpenCV, Scikit-learn can be integrated easily.
I have also selected Keras for preprocessing the word embedding data feature extraction and
training the model on top of Tensor Flow (a library that includes Keras as a sub module).
Pandas is an open-source library that provides high-performance, easy-to-use data structures, and
data analysis tools for the Python programming language[11]. Pandas will be utilized to read the
dataset files and perform different operations on the files.

3.4 Evaluation Technique

Experiments will be conducted to test the functionality of the emotional analysis and detection. To
assess the detection rationally, testing datasets will be put into the analysis. The performance of
the analysis will be evaluated in terms of accuracy, precision, recall, and F-measure.

15
CHAPTER 4: WORK PLAN

4.1 Research Schedule

2022

Task Name Aug Sep Oct Nov Dec Jan

Identification of Problems

Literature Review

Data Collection and Data


analysis

study of various algorithms

Prepare model for training,


validation and testing

Prepare the final document,

4.2 Research Budget

Services Description Qty Price(birr) total Remark

Stationary Stationary material for 200 2000br 2000


material writing, printing and birr
copying the research
papers

Internet Internet access to gather 4mbps for 6 899 899birr


information’s

Other Transportation 2000 birr 2000 2000birr

Total 4899birr

16
References

[1] W. Li, J. E. O’Brien, S. M. Snyder, and M. O. Howard, “Characteristics of Internet


Addiction/Pathological Internet Use in U.S. university students: A qualitative-method
investigation,” PLoS One, vol. 10, no. 2, 2015, doi: 10.1371/journal.pone.0117372.
[2] Y. Lin, X. Wang, and A. Zhou, “Opinion spam detection,” Opin. Anal. Online Rev., no.
May, pp. 79–94, 2016, doi: 10.1142/9789813100459_0007.
[3] Wondwossen Philemon and Wondwossen Mulugeta, A Machine Learning Approach to
Multi-Scale Sentiment Analysis of Amharic Online Posts HiLCoE, Computer
Science Programme.
[4] Mihret, M., & Atinaf, M. (2019). Sentiment Analysis Model for Opinionated Awngi Text.
IEEE AFRICON Conference, 2019-Sep, doi.org/10.1109/AFRICON46755.2019.9134016.
[5] Hiwot Wonago Kululo, “Information filtering of social media Amharic texts based
……. on sentiment Analysis”, 2020.

[6] Megersa Oljira” Rase Sentiment Analysis of Afaan Oromoo Facebook Media Using Deep
……. Learning Approach”, 2020.

[7] Eyob Tesfu “Deep Learning Based Emotion Detection Model for Amharic Text”, 2020

[8] Tocoglu, M. A., Ozturkmenoglu, O., & Alpkocak, A. (2019). Emotion Analysis from
…… Turkish Tweets Using Deep Neural Networks. IEEE Access, 7, 183061–183069. …….
……...https://doi.org/10.1109/ACCESS.2019.2960113.

[9] Park, S. H., Bae, B. C., & Cheong, Y. G. (2020). Emotion recognition from text stories
……..using an emotion embedding model. Proceedings - 2020 IEEE International Conference on
. Big Data and Smart Computing, BigComp 2020, 579–583,
……..https://doi.org/10.1109/BigComp48618.2020.00014.

[10] Karna, M., Juliet, D. S., & Joy, R. C. (2020). Deep learning based Text Emotion
……...Recognition for Chatbot applications. Proceedings of the 4th International Conference on
……..Trends in Electronics and Informatics, ICOEI 2020, Icoei, 988–993,
……..https://doi.org/10.1109/ICOEI48184.2020.9142879.

[11] Wang, H., & Zhao, D. (2020). Emotion analysis of microblog based on emotion dictionary
……...and Bi-GRU. Proceedings of 2020 Asia-Pacific Conference on Image Processing,
……...Electronics and Computers, IPEC 2020, 197–200,
……...https://doi.org/10.1109/IPEC49694.2020.9115194.

17

You might also like