Table of Contents

Abbreviation 2
Base Paper Title 2
Modified Title 2
Modified Title Explanation 2
Abstract 2
Introduction 3
Motivation 5
Objectives 5
Problem Statement 6
Existing System 6
Drawbacks of Existing System 7
. Dataset Desc 7
Proposed System 8
Advantages of Proposed System 8
Hardware & Software Requirements 9
Architecture 10
Existing Algorithm 10
Proposed Algorithm 10
Advantages of Proposed Algorithm 10
Project Modules 11
Literature Survey 13
Conclusion 23
Future Work 23
References 23

Thanks for Choosing WISEN for your needs. Kindly refer your friends to WISEN.
 (c) Wisen IT Solutions Page 2 of 26

Abbreviation

CNN Convolution Neural Network.

LSTM Long Short-Term Memory

Base Paper Title

Multi-Label Emotion Classification on Code-Mixed Text: Data and Methods

Modified Title

Semantic Emotion Analysis: Detecting and Extraction Emotions from Twitter Text using CNN-LSTM

Modified Title Explanation

Semantic : A system for semantic analysis determines the meaning of words in text. Semantics gives a deeper
understanding of the text in sources.
CNN-LSTM : It isa Deep learning Algorithm. We will use this algorithm to detect emotion.

Abstract

Microblog an online-based broadcast medium is a widely used forum for people to share their thoughts and
opinions. Recently Emotion Recognition (ER) from microblogs is an inspiring research topic in diverse areas. In the
machine learning domain automatic emotion recognition from microblogs is a challenging task especially for better
outcomes considering diverse content. Emoticon becomes very common in the text of microblogs as it reinforces
the meaning of content. This study proposes an emotion recognition scheme considering both the texts and
emoticons from microblog data. Emoticons are considered unique expressions of the users emotions and can be
changed by the proper emotional words.
The succession of emoticons appearing in the microblog data is preserved and a D Convolutional Neural Network
(CNN) is employed for emotion classification. The experimental result shows that the proposed emotion
recognition scheme outperforms the other existing methods while tested on Twitter data.
In this study we propose a method that leverages both domain-specific word embeddings and task-specific
features to detect emotions from tweets. Our model employs a convolutional neural network (CNN) and a long
 (c) Wisen IT Solutions Page 3 of 26

short-term memory (LSTM) recurrent neural network which takes word level meta embeddings as inputs and
incorporates contextual embeddings to classify noisy short text.

Introduction

The growth of social media over the last one decade has been tremendous with numbers of people joining
doubling almost on a daily basis. This growth has brought about the need for scalable robust and effective
techniques of managing as well as indexing the content they produce. Anything going on in the world is shared and
communicated through the Internet especially in social media. Social media offers people the chance to interact
comment on events and send instant messages all over the globe without geographical barriers. They include
Facebook YouTube and Twitter amongst others. The social media platforms have opened up many research
opportunities because of the amount of information they possess. This information can be used for many
purposes including things such as prediction and detection of emotions and even as warning systems.
Emotion often refers to a complex state of feeling such as happiness joy anger disgust fear love and hatred that
occurs in physical and psychological changes and has an impact on ones thinking and actions. Emotions can have
a significant impact on peoples lives. Peoples emotions can be expressed in a variety of ways including speech
facial expressions bodily gestures verbal expressions using text and so on . Different social media platforms like
Facebook Twitter Instagram WhatsApp Sina Weibo etc. have grown in popularity in recent years where people
express their emotions and thoughts. Users share millions of tweets and posts every day. Therefore social media
contents are the most prospective sources for understanding emotional states and human thoughts. Microblog an
online-based broadcast medium is a widely used forum for people to share their thoughts and opinions. In our
daily lives microblogs make communication easier. Twitter LinkedIn Facebook Instagram Snapchat Tumblr
WhatsApp etc. are the most popular microblogs. Globally there are . billion users on social media according to
social media projections which equate to approximately % of the current population and this figure is only
increasing. Any post can be hit immediately by a large number of people through microblogs. People express their
thoughts by sharing posts that reflect ones sentiments.
Arguably no other technology has transformed humanity as fundamentally and as quickly as the social web . Social
media platforms like Facebook Twitter Instagram Snapchat and Tiktok now connect billions of people worldwide
enabling them to share status updates images videos and links to online content. The social web democratized the
production and distribution of content reducing the power of traditional gate keepers to decide what information
gets attention; it created a cottage industry of influencersordinary people who have a gift for cultivating online
audiences; it enabled myriads to stay connected to friends around the clock; it catalyzed mass protest movements;
it provided a platform for a digital town square and replaced traditional forms of entertainment.
As the social web matures we are evolving along with it. Despite the growing pains listed above we have become
more comfortable disclosing our inner thoughts and feelings online. As a result we are getting connected to many
others within the emotional web. This too will lead to a profound social transformations. Emotions are a
fundamental part of human experience: they shape how we consume information who we pay attention to what we
believe and how we react to new things . Even without the benefit of visual cues (e.g. facial expressions) audio (e.g.
tone of voice) and other physiological signals the language of text messages conveys an emotional tone. We read
 (c) Wisen IT Solutions Page 4 of 26

other peoples emotions and they affect our own mood. Emotions spread from one person to another
synchronizing feelings of large populations at times erupting as harassment mobs or viral memes
The social web is poised for another transformation. Two major developments make this possible. First although
sites like Twitter and Facebook have long asked people to assess their momentary status with prompts like Whats
happening or Whats on your mind people have grown more comfortable disclosing their private thoughts and
emotional states. The second major development was facilitated by advances in natural language processing.
Language mediates social interactions and captures not only semantics or meaning of conversations but also the
feelings attitudes and even implicit biases that cloud human judgment. New computational tools enable automatic
quantification of a range of emotions expressed in text promising to deliver technology that understands human
experience.
Emotion qualia refers to the raw feel of an emotion. The actual phenomenon of a particular emotion experienced
may differ according to each persons perception or understanding of that emotion with perception being the
result of the individuals past and hypothesised responses unique to each human being. Studies describes the act
of conceptualising core affect or in other words why people attach emotion labels to the experience of emotion
qualia. Since emotion keywords are constructed from conceptual knowledge about the world emotions themselves
may be concepts that humans begin learning in infancy and continuously extend and revise throughout life. This
repeated experience of labelling a combination of core affect and the context in which it occurs as an emotion
provides training in how to recognise and respond to that emotion.
The social web has linked millions of people worldwide creating digital town squares for exchanging ideas opinions
and beliefs. On platforms like Reddit and Twitter among many others people post messages or respond to the
messages posted by others. The low barriers to entry into global online conversations offers society many benefits
such as democratizing the production and distribution of information reducing the power of traditional
gatekeepers to decide what information gets attention creating better ways for people to learn from the diverse
experiences of others and catalyzing mass protest movements. Unfortunately the same mechanisms that lead to
societal benefits are also responsible for creating unique new vulnerabilities. Exchanging diverse viewpoints within
global online communities invites disagreement which malicious actors and anti-social trolls exploit to derail
conversations spread misinformation and inflame polarization. The rise in anti-social online behaviors has had
profound consequences on society undermining collective trust in institutions and in democracy itself .
Online communities have tried to reduce harmful speech by mediating discussions to remove messages that
violate community norms due to toxicity harassment or personal attacks . However manual moderation does not
scale to the volume and speed of online conversations. Although machine moderation has improved in recent
years with tools that automatically recognize harassment hate speech and other types of toxic speech these
methods treat the symptoms rather than causes of the problem. In order to better identify and mediate
controversy we need to understand how controversy develops and derails conversations in open online
communities before we can effectivelyand automatically moderate them.
Early works on understanding emotions relied on dictionary-based methods to measure the sentiment expressed
in messages by counting how many positive or negative words they contained . Researchers found that the
sentiment of tweets in aggregate showed the characteristic diurnal and weekly patterns of mood variation .
Another work found that sentiment tracks the geographic distribution of subjective wellbeing . A new generation of
methods based on large language models enabled a wider range of emotional expressions to be quantified at scale
 (c) Wisen IT Solutions Page 5 of 26

. Depending on training data these models can recognize a dozen or more emotions such as love joy fear anger
disgust etc.
Emotion Recognition (ER) from microblog data is the most promising and challenging research finding in the field
of information and communication technology. Unimodal bimodal and multimodal are the three possible
categories of ER. Unimodal emotion recognition uses only one type of information such as facial expression text or
speech whereas bimodal emotion recognition uses both speech and facial expression. This study introduces the
unimodal emotion recognition method to determine ones Emotion Category (EC). Machine Learning (ML) based
approach and knowledge-based approach are two major techniques for identifying or recognizing emotions. The
knowledge-based technique also known as the lexicon-based technique uses a set of rules to detect emotion from
given data whereas the ML-based technique uses a model for learning the patterns from features generated from
microblog data.
Deep Learning (DL) based approaches for emotion recognition from microblog data have emerged remarkably and
shown promising results nowadays. DL employs different architectures to learn the patterns in data. There are two
major steps in DL-based emotion recognition from microblog data: (i) Processing data and (ii) classifying them
using the proper DL model. Firstly the collected data is processed transformed and represented in the appropriate
format for the envisioned DL model. Secondly a DL model is prepared or trained with the data to classify emotion.
Along with different processing techniques different DL models are investigated in the last several years for
emotion recognition. These DL-based methods run on preprocessed data and do not include explicit functionality.
Some experiments consider emoticons along with the text as input data while others only consider the text.

Motivation

The emoticon and its interactions with texts are given particular consideration in the proposed method. To identify
the true emotion of people both emoticons and text possess equal significance. Predicting causal interaction
between emotions is a complex task because it depends on interactions between the particular linguistic
expression of information semantic context established in the text knowledge of the causal relationships for the
domain in question and the communicative goals of the texts author. This motivates us.

Objectives

 It is challenging to extract complete lexical features and understand causality based on any single
grammatical model.
 How do comments change emotions in the discussion?
 Leads to significant performance improvement.
 Surface meaning usually contradicts its inner, deeper meaning.
 Extracts contextual features and semantic relationships of the words.
 (c) Wisen IT Solutions Page 6 of 26

Problem Statement

Causality understanding between emotions is a critical natural language processing task that is helpful in many
areas including health care business risk management and finance. On close examination one can find a huge
amount of textual content both in the form of formal documents or in content arising from social media like
Twitter dedicated to communicating and exploring various types of causality in the real world. Recognizing these
Cause-Effect relationships between natural language emotions continues to remain a challenge simply because it is
often expressed implicitly. Implicit causality is hard to detect through most of the techniques employed in
literature and can also at times be perceived as ambiguous or vague. Also although well-known datasets do exist
for this problem the examples in them are limited in the range and complexity of the causal relationships they
depict especially when related to implicit relationships. Most of the contemporary methods are either based on
lexico-semantic pattern matching or are feature-driven supervised methods. Therefore as expected these methods
are more geared towards handling explicit causal relationships leading to limited coverage for implicit relationships
and are hard to generalize.

Existing System

The multi-label emotion classification task aims to identify all possible emotions in a written text that best
represent the authors mental state. In recent years multi-label emotion classification attracted the attention of
researchers due to its potential applications in e-learning health care marketing etc. There is a need for standard
benchmark corpora to develop and evaluate multi-label emotion classification methods. The majority of benchmark
corpora were developed for the English language (monolingual corpora) using tweets. However the multi-label
emotion classification problem is not explored for code-mixed text for example English and Roman Urdu although
the code-mixed text is widely used in Facebook posts/comments tweets SMS messages particularly by the South
Asian community. For filling this gap the existing study presents a large benchmark corpus for the multi-label
emotion classification task which comprises code-mixed (English and Roman Urdu) SMS messages. Each code-
mixed (English and Roman Urdu) SMS message manually annotated using a set of emotions including anger
anticipation disgust fear joy love optimism pessimism sadness surprise trust and neutral (no emotion). As a
secondary contribution the existing system applied and compared state-of-the-art classical machine learning
(content-based methods three word n-gram features and eight character n-gram features) deep learning (CNN
RNN Bi-RNN GRU Bi-GRU LSTM and Bi-LSTM) and transfer learning-based methods (BERT and XLNet) on existing
proposed corpus. After our extensive experimentation the best results were obtained using state-of-the-art
classical machine learning methods on word uni-gram (Micro Precision = . Micro Recall = . Micro F = .) with a
combination of OVR multi-label and SVC single-label machine learning algorithms. Our proposed corpus is free and
publicly available for research purposes to foster research in an under-resourced language (Roman Urdu).
The content-based methods for emotion classification tasks are based on n-grams taken from the code-mixed
(English and Roman Urdu) SMS messages. The term n-gram (of characters or words) refers to a series of sequential
tokens in a sentence paragraph and document. A group of n-grams can be generated by considering a series of
tokens moving over a string considering one token at a time. The series can be of length (uni-grams) length (bi-
 (c) Wisen IT Solutions Page 7 of 26

grams) length (tri-grams) etc.

The multi-label emotion classification problem on codemixed SMS messages (English and Roman Urdu) is treated
as a supervised multi-label text classification problem. The existing system applied two different machine learning
classification methods including One vs. Rest and One vs. One along with base classifiers including Random Forest
Logistic Regression Nave Bayes Support Vector Machine Bagging and AdaBoost.
Code-mixed (English and Roman Urdu) text is widely used especially in the South Asian community. However it is
not explored for the multi-label emotion classification problem. As described in this paper our novel contribution is
a newly developed and publicly available benchmark code-mixed and multi-label SMS messages-based corpus for
the multi-label emotion classification task. The corpus consists of code-mixed multi-label SMS messages manually
annotated for the presence/absence of the following emotions: anger anticipation disgust fear joy love optimism
pessimism sadness surprise trust and neutral (no emotion). In addition to dataset creation the existing system
applied state-of-the-art machine learning (content-based -word n-gram features and eight character n-gram
features) deep learning and transfer learning-based methods for multi-label emotion classification task on our
proposed CM-MEC- corpus. The best results (see Table ) obtained using state-of-the-art machine learning methods
with word uni-gram (Micro F = .) and OVR multi-label machine learning algorithm when training on code-mixed
(Train-CM-CM-MEC-) and testing by combining code-mixed Roman Urdu and English (Test-CM-CM-MEC- + Test-E-CM-
MEC- + Test-RUCM-MEC-) multi-label SMS messages.

Drawbacks of Existing System

 Cannot preserves the contextual information along with the polysemy of words.
 The existing model does not focus on the diverse grammatical structures.
 Require extensive knowledge engineering and restrict to a limited domain.
 Cannot improve the diversity and originality of emotion.
 Cannot guarantee the intention of a low similarity between the decoupling representations, which probably
restrains the performance of emotion analysis.

Dataset Desc

Dataset URL: https://aclanthology.org/attachments/D-.Attachment.zip

Dataset Description: Emotions are expressed in nuanced ways which varies by collective or individual
experiences knowledge and beliefs. Therefore to understand emotion as conveyed through text a robust
mechanism capable of capturing and modeling different linguistic nuances and phenomena is needed. We propose
a semi-supervised graph-based algorithm to produce rich structural descriptors which serve as the building blocks
for constructing contextualized affect representations from text. The pattern-based representations are further
enriched with word embeddings and evaluated through several emotion recognition tasks.
 (c) Wisen IT Solutions Page 8 of 26

Proposed System

Controversial comments express more anger and less admiration joy and optimism than non-controversial
comments. Second controversial comments affect emotions of downstream comments in a discussion usually
resulting in long-term increase in anger and a decrease in positive emotions although the magnitude and direction
of emotional change depends on the forum.
The procedure starts with detecting emotions based on text only by using the feature of the bag-of-words which is
calculated using the term frequency-inverse document frequency (TF-IDF) method.
This model is a combination of CNN and LSTM. It stacks two layers of convolution and two LSTM layers then passes
the output to a DNN for prediction. The proposed CNN-LSTM models main benefit is that it considers emoticons in
addition to emotion recognition from real-life Twitter data. Only text data is directed to the proposed CNN-LSTM
method without emoticons and the influence of emoticons in emotion recognition is observed. In any machine
learning system higher test set accuracy is desired because it is the indication of the systems ability for
generalization. More accuracy in the test set specifies that emoticon use in addition to text boosted the capability
of learning the emotion properly of the proposed CNN-LSTM method.

Advantages of Proposed System

 Domain data-distribution helped our model gain performance.

 Language models can learn implicit structural representation in natural language text.
 Identify the relationship along with its semantic features.
 Merges peak phrases into emotion-indicative feature sets.
 Obtain deeper semantics features and attention features of emotional knowledge.
 (c) Wisen IT Solutions Page 9 of 26

Hardware & Software Requirements

Hardware Requirements
 Processor: Minimum i3 Dual Core
 Ethernet connection (LAN) OR a wireless adapter (Wi-Fi)
 Hard Drive: Minimum 100 GB; Recommended 200 GB or more
 Memory (RAM): Minimum 8 GB; Recommended 32 GB or above
Software Requirements
 Python
 Anaconda
 Jupyter Notebook
 TensorFlow
 Keras
 (c) Wisen IT Solutions Page 10 of 26

Architecture

Existing Algorithm

 Gated Recurrent Unit Algorithm

Proposed Algorithm

 CNN-Long Short Term Memory (LSTM) Algorithm

Advantages of Proposed Algorithm

CNN-Long Short Term Memory (LSTM) Algorithm Advantages Advantages

 Explicitly designed to deal with the long-term dependency problem.
 Improved method of back propagating the error.
 Can be used when dealing with large sequences and accuracy is concerned.
 (c) Wisen IT Solutions Page 11 of 26

Project Modules

Module 1 : Text Preprocessing

Certain pre-processing steps are required to eliminate unnecessary content and noisy input from the data. Case
conversion, user name removal, hashtag, punctuation mark removal, and so on are all part of the cleaning
process. Then the clean microblog data containing both texts and emoticons is processed into three processes.
Tokenize :
Convert the string to a list of tokens based on whitespace. This process also removes punctuation marks from the
text.
Stop word filtering :
Eliminate the words which are common and their presence does not tell us anything about the dataset, such as:
the, and, for, etc. Stemming filtering :
Reduce each word to its stem, removing any prefixes or suffixes.
Module 2 : Text Tokenization
Tokenization is the first step in any NLP pipeline. It has an important effect on the rest of your pipeline. A tokenizer
breaks unstructured data and natural language text into chunks of information that can be considered as discrete
elements. The token occurrences in a document can be used directly as a vector representing that document.
Tokenization is a simple process that takes raw data and converts it into a useful data string. While tokenization is
well known for its use in cybersecurity and in the creation of NFTs, tokenization is also an important part of the
NLP process. Tokenization is used in natural language processing to split paragraphs and sentences into smaller
units that can be more easily assigned meaning. The first step of the NLP process is gathering the data (a
sentence) and breaking it into understandable parts (words).
Word tokenizations accuracy is based on the vocabulary it is trained with. These models have to find the balance
between loading words for maximum accuracy and maximum efficiency. While adding an entire dictionarys worth
of vocabulary would make an NLP model more accurate, its often not the most efficient method. This is especially
true for models that are being trained for a more niche purpose.
Module 3 : Model Prediction
We will calculate the vocab_size by using the length extracted from tokenizer.word_index and then add 1 to it. We
are adding 1 because 0 is a reserved for padding and we want to start our count from 1. Finally, we will convert our
predictions data y to categorical data of the vocab size. This function converts a class vector (integers) to the
binary class matrix. This will be useful with our loss which will be categorical_crossentropy. One task in NLP is text
prediction. In a text prediction task, a sentence, or series of words, is presented to a prediction model. The models
task is to then present the word or words with the highest probability of following the initial series to the user.
 (c) Wisen IT Solutions Page 12 of 26

We will be building a sequential model. We will then create an embedding layer and specify the input dimensions
and output dimensions. It is important to specify the input length as 1 since the prediction will be made on exactly
one word and we will receive a response for that particular word. We will then add an LSTM layer to our
architecture. We will give it a 1000 units and make sure we return the sequences as true. This is to ensure that we
can pass it through another LSTM layer. For the next LSTM layer, we will also pass it through another 1000 units
but we dont need to specify return sequence as it is false by default. We will pass this through a hidden layer with
1000 node units using the dense layer function with relu set as the activation. Finally, we pass it through an output
layer with the specified vocab size and a softmax activation. The softmax activation ensures that we receive a
bunch of probabilities for the outputs equal to the vocab size.
 (c) Wisen IT Solutions Page 13 of 26

Literature Survey

Literature Survey 1

Title Sentiment Classification of Crowdsourcing Participants Reviews Text Based on LDA Topic Model

Authors Yanrong Huang , Rui Wang , Bin Huang , Bo Wei , Shu Li Zheng and Min Chen

Published Year 2021

Efficiency  Keeping the control overhead at regular levels

 Reduce resource consumption while meeting reliability demands
 Built-in error handling

Drawbacks  May result in ill-conditioned models.

 Identification of the inconsistencies in the data is very difficult
 Sometimes there is little to chose between competing models.

Description The review text received by crowdsourcing participants contains valuable knowledge, opinions,
and preferences, which is an important basis for employers to make trading decisions, and
crowdsourcing participants to improve service level and quality. However, there are two kinds of
emotional polarity in the review text, the attention paid to sentiment classification of review text
with fuzzy emotional boundaries is insufficient. This paper proposes a supervised text sentiment
classification method with Latent Dirichlet Allocation (LDA) to improve the classification
performance of review text with fuzzy sentiment boundaries. Taking the review text of
crowdsourcing participants on the Zhubajie platform as the data set, using N-gram, Word2vec, and
TF-IDF algorithms to extract text features. The LDA topic model is applied to expand the number
of text features and extract eight topics that affect employers sentiment tendencies. Text
classifiers are constructed based on Support Vector Machine (SVM), Random Forest (RF), Gradient
Boosting Decision Tree (GDBT), and Extreme Gradient Boosting (XGBoost) algorithms, and the
effectiveness of the sentiment classification methods are verified by ten-fold cross-validation and
confusion matrix. Experimental results show that using the LDA topic model to extend the
features of review text can effectively alleviate the problem that the classifier is difficult to
distinguish the sentiment categories of different emotion polarity words coexisting text, and
enhance the ability of emotion boundary fuzzy text classification. Based on TF-IDF and LDA to
extract and expand text features, the GBDT text sentiment classifier with the accuracy of 0.881;
the F1-measure of the second, third, fourth, and fifth categories samples are 0.462, 0.571, 0.278,
and 0.647 respectively, which is better than SVM, RF, and XGBoost classifiers and has the best
classification performance.

In this paper, a supervised text sentiment classication method based on Latent Dirichlet
Allocation(LDA) is pro- posed to improve the classication performance of short text with fuzzy
sentiment boundaries, aiming at more accurately mining the sentiment tendency of users when
two kinds of emotional polarity appear in the review text.
 (c) Wisen IT Solutions Page 14 of 26

Literature Survey 2

Title Classification of Poetry Text Into the Emotional States Using Deep Learning Technique

Authors Shakeel Ahmad , Muhammad Zubair Asghar , Fahad Mazaed Alotaibi and Sherafzal Khan

Published Year 2020

Efficiency  Quick and Efficient to use

 Increasing model skill.
 Better utilization of resources

Drawbacks  Fall into the local optimum, and a long training period.
 Heavyweight
 Problems existing in their respective network structures.

Description The classification of emotional states from poetry or formal text has received less attention by the
experts of computational intelligence in recent times as compared to informal textual content like
SMS, email, chat, and online user reviews. In this study, an emotional state classification system
for poetry text is proposed using the latest and cutting edge technology of Artificial Intelligence,
called Deep Learning. For this purpose, an attention-based C-BiLSTM model is implemented on the
poetry corpus. The proposed approach classifies the text of poetry into different emotional states,
like love, joy, hope, sadness, anger, etc. Different experiments are conducted to evaluate the
efficiency of the proposed system as compared to other state-of-art methods as well as machine
learning and deep learning methods. Experimental results depict that the proposed model
outperformed the baselines studies with 88% accuracy. Furthermore, the analysis of the statistical
experiment also validates the performance of the proposed approach.

To categorize English poetry text within multiple emo- tion classes, we have exploited a deep
learning technique namely, Attention-based C-BiLSTM model. For experimen- tation, a benchmark
dataset is used with an extension into the emotion classes: Alone, Hope, Nature, and Surprise
along with their respective poems. Then this dataset is passed through the following modules, i)
Data Acquisition, ii) Pre- Processing, iii) Feature representation, iv) Feature extraction v) Feature
encoding, vi) Context information generation, and vii) classication. Results depict that the
proposed approach attained highest performance in terms of better (0.88%) pre- cision, (0.88)
Recall, (0.88%) f-measure, and(88%) accuracy, as compared to the state of the art studies. The
limitations of this work include: (i) limited size of the poetry dataset, which is needs to be
increased to get the model trained with improved accuracy, (ii) the system pre- dicts only one
emotion category from 13 classes, which needs to extended to predict multiple emotions with
proper ranking mechanism, and (iii) we used only random word embedding model, whereas, there
is a need to investigate other pre- trained word embedding models. Furthermore, only text- based
features are considered, whereas inclusion of other fea- tures like smiles may produce more
efcient results.
 (c) Wisen IT Solutions Page 15 of 26

Literature Survey 3

Title Adaptive Feature Extractor of Global Representation and Local Semantics for Text Classification

Authors Chaofan Wang , Shenggen Ju , Yuezhong Liu , Run Chen and Xiaoming Huang

Published Year 2020

Efficiency  Outstanding Learning Capabilities

 Lowering the Complexity Threshold
 Provides the integrity and nontransferablity.

Drawbacks  This process not only increases the training time, but also leads to poor stability of the
network.
 Shorten the processing time at the expense of reducing the detection accuracy
 Maximizes the complexity of the problem

Description A hybrid model of Convolutional Neural Network (CNN) and self-attention has achieved
remarkable results in text classification fields. In previous researches, text local
semantics(captured by CNN) and global representation (extracted by self-attention) play equally
important roles for each input. However, the importance of the two varies greatly with complex
linguistic backgrounds. In this paper, we take an adaptive approach to automatically determine the
contribution degree of each model to classification, according to specific structure and grammar
information of the text. This strategy can make the most of two models. To better extract variable-
size features of a word, multi-scale feature attention is introduced into our hybrid model. The
attention focus on assigning larger weights to those multi-scale features that are important to a
word. In addition, for fine-grained emotion classification tasks, a new type of loss function is also
established. Experiment results show that, the proposed model makes noticeable improvements
over hybrid models. Metrics are 0.3 to 1.5 percentages higher than previous methods. And results
also prove that the new loss function further improves the performance of our model in all fine-
grained emotion classification datasets.

In this paper, we present a new hybrid modelMulCNN- Att of CNN and self-attention. Under
complex linguistic backgrounds, MulCNN-Att can adaptively determine which model is more or
less important to classication tasks. In this way, we can maximize the advantages of this two
models. In addition, multi-scale feature attention is also introduced into our hybrid model. The
attention can automatically select task-friendly and effective multi-gram features from texts.
Besides, as for ne-grained emotion tasks, we also establish a new loss function, which aims at
maximizing classica- tion accuracy while minimizing nearest neighbor misclassify- cations.
Experiments results demonstrate that classication accuracy increases by up to 2 percentage point
compared with classic hybrid models based on CNN and self-attention.
 (c) Wisen IT Solutions Page 16 of 26

Literature Survey 4

Title Tree-Structured Neural Networks With Topic Attention for Social Emotion Classification

Authors Chang Wang , Bang Wang and Minghua Xu

Published Year 2019

Efficiency  Improved reliability and resilience

 Works with very high degree of confidence.
 Offer increased flexibility

Drawbacks  Higher prediction complexity with higher dimensions

 This system is Opportunistic and uncontrollable
 More time consuming while the number of network configurations is very large.

Description Social emotion classification studies the emotion distribution evoked by an article among
numerous readers. Although recently neural network-based methods can improve the classification
performance compared with the previous word-emotion and topic-emotion approaches, they have
not fully utilized some important sentence language features and document topic features. In this
paper, we propose a new neural network architecture exploiting both the syntactic information of a
sentence and topic distribution of a document. The proposed architecture first constructs a tree-
structured long short-term memory (Tree-LSTM) network based on the sentence syntactic
dependency tree to obtain a sentence vector representation. For a multi-sentence document, we
then use a Chain-LSTM network to obtain the document representation from its sentences hidden
states. Furthermore, we design a topic-based attention mechanism with two attention levels. The
word-level attention is used for weighting words of a single-sentence document and the sentence-
level attention for weighting sentences of a multi-sentence document. The experiments on three
public datasets show that the proposed scheme outperforms the state-of-the-art ones in terms of
higher average Pearson correlation coefficient and MicroF1 performance.

In this paper, we have proposed hierarchical tree-structured neural networks with LDA attention for
social emotion clas- sication. The lower layer Tree-LSTM network encodes sentence syntactic
information to sentence representation according to the sentence dependency tree analysis. We
have also proposed to include LDA attention to identify document topic-related key words or key
sentences and assign them higher weights when composing document representation.
Experiments on three public datasets show that the proposed schemes with integration of
syntactic and topical informa- tion can effectively improve system performance in terms of higher
MicroF1 and AP in both SSDoc and MSDoc dataset, compared with the state-of-the-art schemes.
In this paper, we have mainly focused on mining the syntactic information within a sentence. In
our future work, we would like to further mine some hidden relations in between sentences, say for
example, by exploring some linguistic knowledge when modeling a paragraph and a document.
 (c) Wisen IT Solutions Page 17 of 26

Literature Survey 5

Title Utilizing External Knowledge to Enhance Semantics in Emotion Detection in Conversation

Authors Fuji Ren and Tianhao She

Published Year 2021

Efficiency  Simplicity and Explainability.

 Loss function is optimized
 Reveal the highly nonlinear Relationship

Drawbacks  Poor Application Performance

 More time consuming while the number of network configurations is very large.
 Sometimes there is little to chose between competing models.

Description Enabling machines to emotion recognition in conversation is challenging, mainly because the
information in human dialogue innately conveys emotions by long-term experience, abundant
knowledge, context, and the intricate patterns between the affective states. We address the task of
emotion recognition in conversations using external knowledge to enhance semantics. We propose
KES model, a new framework that incorporates different elements of external knowledge and
conversational semantic role labeling, where build upon them to learn interactions between
interlocutors participating in a conversation. We design a self-attention layer specialized for
enhanced semantic text features with external commonsense knowledge. Then, two different
networks composed of LSTM are responsible for tracking individual internal state and context
external state. In addition, the proposed model has experimented on three datasets in emotion
detection in conversation. The experimental results show that our model outperforms the state-of-
the-art approaches on most of the tested datasets.

This paper proposes utilizing external knowledge to enhance semantics network architecture that
incorporates con- versational semantic role labeling Information and the commonsense knowledge
feature from ATOMIC for emo- tion recognition in conversation. A knowledge enhanced language
representation layer based on self-attention has been developed for fusion extraction. Based on
the utterance representations rich in external knowledge, the contextual [20] R. Jia and P. Liang,
Adversarial examples for evaluating reading com- prehension systems, in Proc. Conf. Empirical
Methods Natural Lang.
 (c) Wisen IT Solutions Page 18 of 26

Literature Survey 6

Title TBLC-rAttention: A Deep Neural Network Model for Recognizing the Emotional Tendency of
Chinese Medical Comment

Authors Qibing Jin , Xingrong Xue , Wenjuan Peng , Wu Cai , Yuming Zhang and Ling Zhang

Published Year 2020

Efficiency  Reduction in features to improve the quality of prediction.

 Scalable training framework well balances the estimation.
 Relatively simple and computationally inexpensive method

Drawbacks  Unsuitable for large scale scenarios.

 Poor performance and high variance on the validation data.
 Cannot be implemented real time

Description In the current paper, a hybrid depth neural network model, TBLC-rAttention, aiming at Chinese text
emotion recognition, is proposed to identify the emotional tendency of the Chinese medical
reviews. The model includes the following steps: acquiring and preprocessing the Chinese corpus;
mapping the preprocessed text into the word vectors; using Bi-directional Long Short-Term
Memory network (Bi-LSTM) with the attention mechanism to acquire the context semantic
features of the text; using Convolutional Neural Network (CNN) to obtain local semantics features
on the basis of the context semantic features; and inputting the final feature vectors into the
classification layer to complete the task of emotion recognition and the classification of the
Chinese medical reviews. In this experiment, the corpus data is the comments of 999 cold
medicine on a large e-commerce platform. All corpus are divided into three types, including high
praise, medium praise and bad review. Classical machine learning models (SVM, NB) and neural
network models (CNN, LSTM, Bi-LSTM, BiLSTM-Attention and RCNN) are performed as the
comparison benchmarks to assess the category performance of TBLC-rAttention model. All the
results were obtained when the training accuracy and test accuracy were stable after 1000 cycles
of repeated calculation. The results show that TBLC-rAttention can get better text feature than the
reference models, and the text classification accuracy reaches to 99%. In conclusion, the TBLC-
rAttention model can identify semantic feature information to the greatest extent. In addition, this
study also completes the numerical quantification of the predicted results.

In this paper, a hybrid deep neural network model, TBLC- rAttention, aiming at Chinese text
emotion recognition, is proposed. This model can be used for knowledge extraction, modeling,
classication and mining of the medical texts. In this study, the performance of the TBLC-rAttention
is evaluated and veried on the collected data set. Although range. Deep learning is an effective
technique for analyzing and processing medical texts. Although this study has made some
achievements in this respect, the classication and mining of medical text using models with better
performance and higher intelligence level remains a challenging eld.
 (c) Wisen IT Solutions Page 19 of 26

Literature Survey 7

Title A Self-Attentive Convolutional Neural Networks for Emotion Classification on User-Generated

Contents

Authors Ying Qian , Weiwei Liu and Jiangping Huang

Published Year 2020

Efficiency  Powerful Reprsentation Capability

 Reduce resource consumption while meeting reliability demands
 Quick and Efficient to use

Drawbacks  Cant learn relation between factors.

 Unsuitable for large scale scenarios.
 Fail to capture the hidden patterns, their prediction performance is relatively poor.

Description It is challenging to detect emotions on user-generated contents (UGC) because it tends to have
sparse emotional semantics, multiple emotions in the same text, and a fast update of emotional
expression. Word embedding can extract high-level features in words to enrich their semantics.
The convolutional neural networks (CNNs) can make model training more efficient to adapt to
fast-changing emotional expressions, but its original pooling operation can not simultaneously
filter out multiple emotional features that are beneficial to classification results, which is not
suitable for UGC. In this paper, we propose a self-attentive convolutional neural networks
(SACNNs) trained on top of pre-trained word vectors for emotion detection on UGC, which reserves
different kinds of emotion information, avoids the loss of emotional aspects in the pooling
process of CNNs structure, and increases the interpretability of the model by visualizing the
extraction process of features. Meanwhile, the convergence speed of the model training is
accelerated, and the model is updated in real time to detect emotions with rich and novel
expressions. The proposed model combines the CNNs and the self-attention mechanism, the self-
attention mechanism can select key emotional features after convolution. We evaluate the
proposed model with two datasets from NLPCC 2014 and SemEval 2018 Task 1. The
experimental results show that our model obtain significant performance than baselines in multi-
label classification on UGC. In addition, the experimental results and the rationality of the self-
attention mechanism are analyzed in detail, and the influential convolutional filter windows are
visualized based on attention weights.

In this paper, we propose a SACNNs model, which combines CNNs model with the self-attention
mechanism. That is, the self-attention mechanism is used to replace the pooling layer in the
traditional CNNs, and the key local seman- tic feature information is effectively selected and
merged. Finally, our model achieves higher Ap, Micro-F and Macro-F in the classic text classication
methods, and shows bet- ter robustness. In addition, the lter size and convolutional layer of the
CNNs can be adjusted and changed as needed, which indicates its powerful generalization and
extensibility. In addition, we also detail the main explanations of examples, visualization and
quantitative analysis.
 (c) Wisen IT Solutions Page 20 of 26

Literature Survey 8

Title Variable Convolution and Pooling Convolutional Neural Network for Text Sentiment Classification

Authors Min Dong , Yongfa Li , Xue Tang , Jingyun Xu , Sheng Bi and Yi Cai

Published Year 2020

Efficiency  Simple, fast and less complex.

 Powerful Reprsentation Capability
 High Effective with Complex Problems

Drawbacks  Decrease the generalizability of the model.

 Cannot reduce the variance of predictions; therefore cannot reduce generalization error.
 Poor Application Performance

Description With the popularity of the internet, the expression of emotions and methods of communication are
becoming increasingly abundant, and most of these emotions are transmitted in text form. Text
sentiment classification research mainly includes three methods based on sentiment dictionaries,
machine learning and deep learning. In recent years, many deep learning-based works have used
TextCNN (text convolution neural network) to extract text semantic information for text sentiment
analysis. However, TextCNN only considers the length of the sentence when extracting semantic
information. It ignores the semantic features between word vectors and only considers the
maximum feature value of the feature image in the pooling layer without considering other
information. Therefore, in this paper, we propose a convolutional neural network based on multiple
convolutions and pooling for text sentiment classification (variable convolution and pooling
convolution neural network, VCPCNN). There are three contributions in this paper. First, a
multiconvolution and pooling neural network is proposed for the TextCNN network structure.
Second, four convolution operations are introduced in the word embedding dimension or direction,
which are helpful for mining the local features on the semantic dimensions of word vectors.
Finally, average pooling is introduced in the pooling layer, which is beneficial for saving the
important feature information of the extracted features. The verification test was carried out on
four emotional datasets, including English emotional polarity, Chinese emotional polarity, Chinese
subjective and objective emotion and Chinese multicategory. Our approach is effective in that its
result was up to 1.97% higher than that of the TextCNN network.

A multiconvolution and pooling method for text sentiment classication is proposed in this paper,
which is based on the TextCNN network structure. For better text feature extraction, the model
introduces four different convolution operations in the word embedding dimension and adds
average pooling in the pooling layer to extract more detailed local features. The experimental
results show that compared with methods such as DCNN, MVCNN, RCNN and Bi-LSTM, our
method is greatly improved, especially in the Chinese multicate- gory dataset. The VCPCNN-
2D_DIFF structure is 14.60% higher than the MVCNN method, although the VCPCNN- 2D_SAME
structure decreases less than 1% in the English emotional polarity dataset compared with RCNN.
 (c) Wisen IT Solutions Page 21 of 26

Literature Survey 9

Title Emotion Recognition by Textual Tweets Classification Using Voting Classifier (LR-SGD)

Authors Anam Yousaf , Muhammad Umer , Saima Sadiq , Saleem Ullah , Seyedali Mirjalili , Vaibhav
Rupapara and Michele Nappi

Published Year 2021

Efficiency  Ensure the reliability of the observed linear relationship.

 Better utilization of resources
 Provides more flexibility when designing models.

Drawbacks  High prediction complexity for large datasets

 Problem of Diminishing Feature Reuse
 Unpredictability in the training is higher.

Description The proliferation of user-generated content on social media has made opinion mining an arduous
job. As a microblogging platform, Twitter is being used to collect views about products, trends,
and politics. Sentiment analysis is a technique used to analyze the attitude, emotions and
opinions of different people towards anything, and it can be carried out on tweets to analyze
public opinion on news, policies, social movements, and personalities. By employing Machine
Learning models, opinion mining can be performed without reading tweets manually. Their results
could assist governments and businesses in rolling out policies, products, and events. Seven
Machine Learning models are implemented for emotion recognition by classifying tweets as
happy or unhappy. With an in-depth comparative performance analysis, it was observed that
proposed voting classifier(LR-SGD) with TF-IDF produces the most optimal result with 79%
accuracy and 81% F1 score. To further validate stability of the proposed approach on two more
datasets, one binary and other multi-class dataset and achieved robust results.

This paper proposed a novel combination of LR and SGD as a voting classier for emotion
recognition by classify- ing tweets as happy or unhappy. Our experiments showed that one can
improve the performance of models by recog- nizing patterns efciently and through effective
averaging combination of models. Experiments are conducted to test seven machine learning
models that are; (1) SVM, (2) RF, (3) GBM, (4) LR, (5) DT, (6) NB and (7) VC(LR-SGD). This study
also employed two feature representation techniques Tf and TF-IDF. The results showed that all
models performed well on tweet dataset but our proposed voting classier VC(LR-SGD) outperforms
by using both TF and TF-IDF among all. Proposed model achieves the highest results using TF-
IDF with 79% Accuracy, 84% Recall and 81% F1-score. The proposed model is further validated on
two more dataset and achieved robust results. The future work will compare more feature
engineering techniques and explore more com- binations of ensemble models to improve the
performance. In addition, new techniques will be investigated to deal with sarcastic comments.
 (c) Wisen IT Solutions Page 22 of 26

Literature Survey 10

Title Extracting Emotion Causes Using Learning to Rank Methods From an Information Retrieval
Perspective

Authors Bo Xu , Hongfei Lin , Yuan Lin , Yufeng Diao , Liang Yang and Kan Xu

Published Year 2019

Efficiency  Minimizes the workload on infrastructures.

 Maximum Utilization of Unstructured Data
 Boost the Performance

Drawbacks  Face challenges in both accuracy and memory

 Solutions have been proved ineffective
 Cannot reduce the variance of predictions; therefore cannot reduce generalization error.

Description Emotion cause extraction is a challenging task for the fine-grained emotion analysis. Even though
a few studies have addressed the task using clause-level classification methods, most of them
have partly ignored emotion-level context information. To comprehensively leverage the
information, we propose a novel method based on learning to rank to identify emotion causes
from an information retrieval perspective. Our method seeks to rank candidate clauses with
respect to certain provoked emotions in analogy with query-level document ranking in information
retrieval. To learn effective clause ranking models, we represent candidate clauses as feature
vectors involving both emotion-independent features and emotion-dependent features. Emotion-
independent features are extracted to capture the possibility that a clause is expected to provoke
an emotion, and emotion-dependent features are extracted to capture the relevance between
candidate cause clauses and their corresponding emotions. We investigate three approaches to
learning to rank for emotion cause extraction in our method. We evaluate the performance of our
method on an existing dataset for emotion cause extraction. The experimental results show that
our method is effective in emotion cause extraction, significantly outperforming the state-of-the-art
baseline methods in terms of the precision, recall, and F-measure.

In this paper, we propose a novel clause ranking method to tackle the problem of emotion cause
extraction using learning to rank. We rst transform emotion cause extraction as a supervised
ranking problem from an information retrieval perspective. We then dene and extract a large
amount of emotion-independent and emotion-dependent clause features for emotion cause-
oriented clause representations. We inves- tigate three approaches to learning to rank for
constructing clause-level ranking models. We evaluate our models using a publicly available
dataset for emotion cause extraction. Experimental results show that the proposed models is
effec- tive in identifying emotion causes. Our models outperform B. Xu et al.: Extracting Emotion
Causes Using Learning to Rank Methods From an Information Retrieval Perspective the state-of-
the-art models in terms of the precision, recall and F-measure. In the future, we will construct more
powerful ranking models by developing effective ranking features for emotion cause extraction.
 (c) Wisen IT Solutions Page 23 of 26

Conclusion

Technology is changing us. The social web (in concert with smartphones) has already altered how we interact with
others and how we spend our time. Our transformation is still in its nascent stages. The next step of the
development will involve creating emotional connections that enable us to share our loves and joys but also anger
disappointment and disgust. In parallel we are creating tools that can better recognize the emotions in text but
soon also in images and videos. This will allow us to build technologies that understand people on a deeper
psychological level that can read our emotions and adapt to our moods. @@ Nowadays social media has become
the most prevalent podium to express one's feelings & emotions and for a better kind emoticons are commonly
used with texts. In the ML domain emotion recognition from microblog data has emerged as a challenging and
promising research finding. For emotion recognition most of the existing methods consider only text data for
simplicity which is not sufficient. In this study an emotion recognition model using CNN-LSTM is developed
considering emoticons in addition to text. As emoticons can have a significant role in the emotional behavior of
human beings using microblog data the proposed CNN-LSTM technique outperforms other emotion recognition
methods considering emoticons in addition to text using real-life Twitter data. In summary this research developed
an emotion classification technique and the effectiveness of emoticon consideration in emotion recognition from
microblog data.

Future Work

In the future we may explore the following directions: () currently the detected key emotions only have several
documents and an estimated emotion time so it would be interesting to automatically extract their actual
times/locations and name the emotions; () by grouping documents into key emotions we may investigate how to
generalize emotion mention schema across different emotion types; () we can also further model the inner
structure of key emotions as episodes (sub-emotions) to construct the entire emotion structure hierarchy.

References

 P. Yan, W. X. Lu, and J. Su, Construction method of emotion dictionary Jan. 2019.
 Y. W. Wu, K. Huang, and X. Y. Wang, Short text sentiment classication Jun. 2019. 2001.
 O. Alsharif, D. Alshamaa, and N. Ghneim, Emotion classication in p. 16, May 2013.
 S. Soumya, S. Saju, R. Rajan, and N. Sebastian, Poetic meter classi- cation using TMS320C6713 DSK, in Proc.
Int. Conf. Signal Process.
 C. Jareanpon, W. Kiatjindarat, T. Polhome, and K. Khongkraphan, Auto- matic lyrics classication system using
text mining technique, in Proc.
 A. Severyn and A. Moschitti, Twitter sentiment analysis with deep con- volutional neural networks, in Proc.
38th Int. ACM SIGIR Conf. Res.
 (c) Wisen IT Solutions Page 24 of 26

 Y. Zhang, J. Wang, and X. Zhang, YNU-HPCC at SemEval-2018 task 1: BiLSTM with attention based sentiment
analysis for affect in tweets, in the Mathematical Formula(Equation) for Deep Learning Model? Accessed Dec.
 C. Zhou, C. Sun, Z. Liu, and F. C. M. Lau, A C-LSTM neural network for text classication, 2015, arXiv:1511.08630.
Accessed: Feb. 6, 2020.
 Anaconda. Accessed: Nov.

D. Zhang and W. S. Lee, Question classication using support vector machines, in Proc. 26th Annu. Int. ACM
SIGIR Conf. Res. Develop. Inf.
 R. Johnson and T. Zhang, Deep pyramid convolutional neural networks for text categorization, in Proc. 55th
Annu. Meeting Assoc. Comput. [10] S. Wang, M. Huang, and Z. Deng, Densely connected CNN with multi- scale
feature attention for text classication, in Proc. 27th Int. Joint Conf.
 J. Pennington, R. Socher, and C. D. Manning, Glove: Global vectors for word representation, in Proc. Conf.
Empirical Methods Natural Lang.
 R. Collobert, Natural language processing (almost) from scratch, J.

 R. Wang, Z. Li, J. Cao, T. Chen, and L. Wang, Convolutional recurrent neural networks for text classication, in
Proc. Int. Joint Conf. Neural [30] Z. C. Lipton, J. Berkowitz, and C. Elkan, A critical review of recur- rent neural
networks for sequence learning, 2015, arXiv:1506.00019.
 Y. Xiao and K. Cho, Efcient character-level document classication by combining convolution and recurrent
layers, 2016, arXiv:1602.00367.
 Y. Rao, Contextual sentiment topic model for adaptive social emotion C. Wang et al.: Tree-Structured Neural
Networks With Topic Attention for Social Emotion Classification [10] Y. Rao, H. Xie, J. Li, F. Jin, F. L. Wang, and Q.
Li, Social emotion classication of short text via topic-level maximum entropy model, Inf.
 T. Mikolov, I. Sutskever, K. Chen, G. S. Corrado, and J. Dean, Distributed representations of words and
phrases and their compositionality, in Proc.
 J. Pennington, R. Socher, and C. Manning, Glove: Global vectors for word representation, in Proc. Conf.
Empirical Methods Natural Lang.
 X. Li, Y. Rao, H. Xie, X. Liu, T.-L. Wong, and F. Lee Wang, Social emotion classication based on noise-aware
training, Data Knowl. Eng., to be published.
 R. Socher, D. Chen, C. D. Manning, and A. Ng, Reasoning with neural tensor networks for knowledge base
completion, in Proc. Adv. Neural Inf.
 M. Huang, Q. Qian, and X. Zhu, Encoding syntactic knowledge in neural p. 26, 2017.

 F. Ren, From cloud computing to language engineering, affective comput- 2010. [10] C.-C. Lee, E. Mower, C.
Busso, S. Lee, and S. S. Narayanan, Emotion recognition using a hierarchical binary decision tree approach, in
Proc.
 N. Majumder, S. Poria, D. Hazarika, R. Mihalcea, A. Gelbukh, and E. Cambria, DialogueRNN: An attentive RNN
for emotion detection in conversations, 2018, arXiv:1811.00405.
 J. Chung, C. Gulcehre, K. Cho, and Y. Bengio, Empirical evalua- tion of gated recurrent neural networks on
sequence modeling, 2014, arXiv:1412.3555.
 L. He, K. Lee, M. Lewis, and L. Zettlemoyer, Deep semantic role label- ing: What works and whats next, in Proc.
 (c) Wisen IT Solutions Page 25 of 26

55th Annual Meeting Assoc.

 C. Busso, M. Bulut, C.-C. Lee, A. Kazemzadeh, E. Mower, S. Kim, J. N. Chang, S. Lee, and S. S. Narayanan,
IEMOCAP: Interactive emo- p. 335, Dec. 2008.
 Z. Zhang, X. Han, Z. Liu, X. Jiang, M. Sun, and Q. Liu, ERNIE: Enhanced language representation with
informative entities, in Proc. 57th Annu.
 A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. U. Kaiser, and I. Polosukhin, Attention
is all you need, in Proc. Adv. [40] S. Poria, D. Hazarika, N. Majumder, G. Naik, E. Cambria, and R. Mihalcea,
MELD: A multimodal multi-party dataset for emotion recognition in conversations, 2018, arXiv:1810.02508.
 W. Shen, J. Chen, X. Quan, and Z. Xie, DialogXL: All-in-one XLNet for multi-party conversation emotion
recognition, 2020, arXiv:2012.08695.
 W. Shen, S. Wu, Y. Yang, and X. Quan, Directed acyclic graph network for conversational emotion recognition,
2021, arXiv:2105.12907.
 M. C. Daz-Galiano, M. T. Martn-Valdivia, and L. A. Urea-Lpez, Query expansion with a medical ontology to
improve a multimodal infor- Apr. 2009.
 S. R. Das and M. Y. Chen, Yahoo! For amazon: Sentiment extraction Sep. 2007. [10] S. Hassanpour, C. P.
Langlotz, T. J. Amrhein, N. T. Befera, and M. P. Lungren, Performance of a machine learning classier of knee
MRI reports in two large academic radiology practices: A tool to estimate Apr. 2017.
 B. E. Chapman, S. Lee, H. P. Kang, and W. W. Chapman, Document-level classication of CT pulmonary
angiography reports based on an extension Oct. 2011.
 X. Cheng, X. Yan, Y. Lan, and J. Guo, BTM: Topic modeling over short Dec. 2014.

 G. E. Hinton and R. R. Salakhutdinov, Reducing the dimensionality of Jul. 28, 2006.

 Y. Kim, sentence classica- tion, 2014, arXiv:1408.5882. abs/1408.5882
 A survey of recent advances in texture representation, Tech. Rep., 2018.
 A. Joulin, Bag of tricks for efcient text classication, in Proc. Conf.
 R. E. Schapire and Y. Singer, BoosTexter: A boosting-based system May 2000.
 J. S. Y. Liew, H. R. Turtle, and E. D. Liddy, EmoTweet-28: A ne-grained emotion corpus for sentiment analysis, in
Proc. 10th Int. Conf. Lang.
 X. Hao, Y. Jia, and Q. Gu, An automatic construction approach for sentiment dictionary based on weibo
emoticons, in Proc. Int. Conf. Netw., Commun., Comput. Eng. (NCCE), May 2018.
 N. Arunachalam, S. J. Sneka, and G. Madhumathi, A survey on text classication techniques for sentiment
polarity detection, in Proc. Innov.
 J. Song, K. T. Kim, B. J. Lee, S.-Y. Kim, and H. Y. Youn, A novel clas- sication approach based on Nave Bayes for
Twitter sentiment analysis, [20] F. Wang, X. Deng, and L. Hou, Chinese news text multi classication based on
naive bayes algorithm, in Proc. 2nd Int. Symp. Comput. Sci. Intell. Control, 2018, p. 31.
 M. A. Hearst, S. T. Dumais, E. Osman, J. Platt, and B. Scholkopf, Support Jul./Aug. 2008.

 Q.-H. Vo, H.-T. Nguyen, B. Le, and M.-L. Nguyen, Multi-channel LSTM- CNN model for Vietnamese sentiment
analysis, in Proc. 9th Int. Conf.
 (c) Wisen IT Solutions Page 26 of 26

 Y. Zhang, Z. Zhang, D. Miao, and J. Wang, Three-way enhanced convo- lutional neural networks for sentence-
level sentiment classication, Inf.
 A. Alsaeedi and M. Zubair, A study on sentiment analysis techniques of 2019.

 A. M. Arbib, The Handbook of Brain Theory and Neural Networks, 2nd ed. Cambridge, MA, USA: MIT Press,
2002.
 Z. S. Li and A. Jain, Encyclopedia of Biometrics. Berlin, Germany: Springer, 2015.

 W. Gao, S. Li, S. Y. M. Lee, G. Zhou, and C.-R. Huang, Joint learning on sentiment and emotion classication, in
Proc. 22nd ACM Int. Conf. Conf.
 L. Gui, J. Hu, Y. He, R. Xu, Q. Lu, and J. Du, A question answering approach to emotion cause extraction, in
Proc. Conf. Empirical Methods [10] S. Y. M. Lee, Y. Chen, and C.-R. Huang, A text-driven rule-based system for
emotion cause detection, in Proc. NAACL HLT Workshop Comput.
 D. Ghazi, D. Inkpen, and S. Szpakowicz, Detecting emotion stimuli in emotion-bearing sentences, in Proc. Int.
Conf. Intell. Text Process.
 D. Ghazi, D. Inkpen, and S. Szpakowicz, Detecting emotion stimuli in emotion-bearing sentences, in Proc. Int.
Conf. Intell. Text Process.