https://doi.org/10.31449/inf.v46i7.4280                                                                    Informatica 46 (2022) 131-144   131 
 
A Multi-label Classification of Disaster-Related Tweets with 
Enhanced Word Embedding Ensemble Convolutional Neural 
Network Model 
E. Arathi
1
*, S.Sasikala
1 
 
E-mail: aarthi.devpal@gmail.com, sasikalarams@gmail.com 
1 Computer Science, IDE, University of Madras Chennai, Tamil Nadu, India. 
* Corresponding author 
 
Keywords: Twitter data classification, Social Media, Convolutional Neural Network (CNN), Ensemble Deep 
Learning, Embeddings from Language Models (ELMo),  Recurrent Neural Network (RNN),  multi-label classification. 
Received: August 8, 2022 
 
Abstract: 
Recently, adopting machine learning techniques to automate the identification and classification of 
event-related tweets has been beneficial in times of crisis. Word embeddings are the most effective word 
vectors for NLP processing using deep learning classifiers. This research proposes a novel method with 
the Enhanced Embedding from Language Model (EnELMo) for classifying tweets as different categories 
with higher classification accuracy and precision for the rapid rescue action in the disaster scenario. 
The proposed Enhanced Word Embedding Ensemble Convolutional Neural Network Model(EWECNN) 
method comprises an Enhanced ELMo module with crisis lexicon to create Crisis word vectors, a Novel 
ELMo-CNN Architecture module for feature extraction (ECA), and an effective multi-label classification 
of text using Crisis Word Vector specific CNN-RNN(CWV-CRNN) stacks. These functional modules are 
intended to improve the classification. Among the various approaches discussed, the proposed method 
outperforms the classification of microblog texts with an accuracy of 93.46 percent and the  F1-Score of 
92.99 percent for multi-classification of tweets, which is higher than other methods discussed in this 
study. The proposed multi-label classification of disaster-related text facilitates faster rescue action in a 
crisis scenario. 
Povzetek: Predstavljena je nova metoda za iskanje uspešnega reševanja v primeru katastrof na osnovi 
analiz socialnih omrežij z ansamblom nevronskih mrež (EWECNN). 
 
1   Introduction 
Disasters have an impact on societies all over the world 
because they can strike without warning and cause 
massive damage. The reaction and recovery of 
communities during all disaster phases rely on their level 
of preparedness [1]. During a crisis, those affected, 
authorities, and volunteers seek actionable information to 
aid in damage restoration and rescue operations. Timely 
and accurate information is critical for humanitarian and 
government efforts to save lives and gain access to those 
affected [2]. The phenomenal utilization and additional 
accomplishments make the social media platform widely 
used in crisis dissemination.  
Recent advances in Artificial Intelligence have 
resulted in the development of machine learning 
techniques to aid in automating Twitter data multi-
classification. Text classification is the process of 
automatically assigning tags to predefined classes based 
on the content of texts. Different types of tweets are 
posted during a disaster, such as rumors and spam, 
emotional information, and other types of data that need 
to be classified by NLP. In conventional methods, 
different features are hand-crafted to classify tweets of a 
specific type in times of crisis. Various deep learning 
neural network algorithms, such as CNN and RNN, 
achieve high accuracy in crisis-related multi-
classification challenges [4][5][6]. However, more 
research and inventions are required to improve social 
media text data classifications. Data collection, 
preprocessing, feature selection, Construction of a new 
classification methodology, Training, hyper-parameter 
fine-tuning, and evaluation of the proposed method are 
the customary phases of a text classification system [7]. 
The microblog messages are categorized under eight 
categories: Disaster kind, Location, Dead and Injured 
People, Help Request, Infrastructure damage, Search and 
Rescue, Weather-related information, and Non-relevant 
information to reach victims of disasters more quickly 
and efficiently. Using a context-specific Language Model 
and domain-specific ensemble CRNN, a new multi-label 
text classifier is developed in this work. Extensive 
research is conducted for the multi-label categorization 
of tweets with standard CrisisNLP datasets of different 

132 Informatica 46 (2022) 131–144                                                                                                                             E. Arathi et al.
  
disaster events, evaluated using performance metrics. 
This study compares various multi-classification 
approaches with the proposed method. 
 
This paper is structured as follows: The second section 
covers relevant studies on classifying Twitter messages 
for various goals. The third section presents an enELMo 
embedding model employing the ensemble CNN-RNN 
stack classifier. The experiment is described in section 
four. The fifth section elaborates on the model's results 
and analysis of its performance. Finally, Section 6 
concludes the paper. 
 
2   Related Works 
2.1 Crisis-related Tweets Classification 
Recent studies have shown the increasing importance of 
social media during emergencies and how broadcasting 
information via social media can improve situational 
awareness during a crisis. These authors  [8,9,10] were 
the first to look into and analyze the use of microblogs 
and information lifecycles during crisis scenarios. They 
studied the behavior of microbloggers by conducting a 
qualitative analysis of tweets published during a flooding 
incident. The authors of [11] employed the bags-of-
words method to locate a crisis's data. Various 
grammatical elements, like Parts-Of-Speech (POS), are 
influenced by the vocabulary used in Twitter posts. In the 
proposed work, an Artificial Intelligence Disaster 
Response (AIDR) system was developed using uni-gram 
and bi-gram features to classify crisis-related information 
during a disaster. The authors of [12] used the 
informative terms from tweets as characteristics to 
recognize resource tweets during a disaster. Based on 
low-level features of lexical and syntactic characteristics, 
models for disaster situational awareness were developed 
in [13,14]. 
The process of automatically extracting information from 
tweets has brought much attention to word embedding. 
The work proposed in [15, 16] used CNN with word 
embedding and demonstrated superior performance to 
hand-crafted features in disaster-related tasks. In the 
study [17,18], CNN and MLP-CNN with word 
embedding are used to categorize the data linked to 
crises. The    skip-gram model of the word2vec tool was 
applied [19, 20, 21] to extract information from an 
extensive corpus consisting of almost 57,908 tweets. 
During the disaster event, the authors of [22] developed a 
deep learning model using the word embedding to detect 
informative tweets related to the catastrophe for speedier 
actions. 
2.2 Multi-Label Classifications of tweets 
A set of comparable multi-label text classification 
methods are taken here to juxtapose the performance of 
the proposed method. They are, A new big data approach 
for topic classification and sentiment analysis of Twitter 
data (BDACSA)[20], A pattern-based approach for 
multi-class sentiment analysis in Twitter(PAMSA) [23], 
and improved classification of crisis-related data on 
Twitter using contextual representations (CRICCD [24], 
Multi-level aspect-based sentiment classification of 
Twitter data: using the hybrid approach in deep learning 
(MASC) [25] and Sentiment classification from multi-
class imbalanced Twitter data using binarization (BMSC) 
[26]. This section confers a limpid analysis of the 
methodologies used and the advantages and limitations 
of these existing methods. The Naive Bayes classifier 
(HL-NBC) proposed by the authors of [27] can be used 
to classify microblog text into several categories and 
filter out irrelevant tweets. This experiment tested the 
Naive Bayes classifier model with Lexicon, unigram, and 
bigram features. The HL-NBC method improves 
sentiment classification and achieves an accuracy of 82 
percent; however, processing time and cross-lingual 
categorization are limitations, and the context of the text 
cannot be found effectively using this approach.   
SENTA, a user-friendly tool for classifying text 
from microblogging websites into seven distinct 
sentimental categories, was presented in this paper [28]. 
A customizable feature selection option was designed to 
extract the features such as Sentiment features, 
Punctuation features, Syntax/Stylistic Features, and 
semantic features to preprocess the text for classification. 
Compared to the text's binary and ternary classification, 
this model gives a multi-class classification rate of 
60.2%, while the binary and ternary classification rates 
are both 80.1%. To the advantage of this work, there is 
an abundance of configurable feature classifications. 
However, this method's average accuracy for multi-class 
classification is a noted limitation. 
The authors of [29] suggested that the intense 
classifier with ELMo embedding's model achieved a 
higher metric in terms of performance than the regular 
Support Vector Machine (SVM) for the multi-
classification of disasters-related tweets. To arrive at this 
conclusion, the researchers used datasets such as those 
gathered from multiple sources like CrisisNLP, 
CrisisLex, and AIDR Twitter standard datasets from the 
earthquakes in Nepal and California and Typhoon 
Hagupit. An embedding layer can generate word vectors 
from input text. The dense classifier then accurately 
predicts the crisis and other tweet texts by processing the 
word vectors at 82.3 percent. The drawbacks are the 
necessity of contextual feature selection and better 
processing time.  
Using a new classifier MuLeHyABSC and a 
feature ranking process, this study [30] claims to perform 
classification of Twitter data based on multi-level aspect-
based classification. MASC employs an Artificial Neural 
Network Multilayer Perceptron (ANMP) to improve 
classification performance. Several existing machine 
learning classifiers are combined with MuLeHyABSC. 
The evaluation process used the benchmark datasets 
STC, TAS, FGD, ATC, and STS.  the performance of the 
MASC method secured higher scores in terms of 
Accuracy, Precision, Recall, and F-Score. Higher 
classification accuracy and precision are the proven 
advantages of this method. At the same time, the 
increased processing time is the limitation of the MASC 

A Multi-label Classification of Disaster-Related Tweets…                                      Informatica 46 (2022) 131-144   133 
method. The higher processing time makes the 
application  MASC lag in the real-time classification 
processes.  
The authors of [31] presented the proposed methodology 
as one-vs-one binary decomposition and dimension 
reduction. In addition to Weighted multi-class 
reconstruction, a stable preprocessing method was 
introduced to detect minority classes with the MBSC 
method. SemEval2016 Message Polarity classification 
dataset was used in the MBSC method. The modules of 
MBSC were tested with the standard baseline classifiers. 
The performance metric of MBSC work was measured in 
terms of Geometric Mean for multi-class classification. 
Better classification performance is the advantage of this 
method. Table 1 provides a summary of the related 
methodologies, their limitations, and the benefits of the  
existing methods.
 
Table 1: An  outline of the  methodologies, advantages, and limitations of the existing methods 
Author Work Methodology Accuracy 
% 
Advantages Limitations 
Anisha. 
Rodrigues et.al. 
A new big data approach for 
topic classification and 
sentiment analysis of Twitter 
data[20] 
Hybrid Lexicon-
Naïve Bayesian 
Classifier 
82 Processing 
time 
Moderate 
Accuracy 
Mondher 
Bouaziz et.al. 
A pattern-based approach for 
multi-class sentiment 
analysis in Twitter .[21] 
Tokenization, 
Lemmatization, and 
Generating negation 
vectors 
80.1 Highly 
configurable 
Average 
Accuracy  
Sreenivasulu 
Madichetty et.al 
Improved classification of 
crisis-related data on Twitter 
using contextual 
representations[22] 
CNN dense classifier 
with ELMo 
embedding for 
feature extraction. 
82.3 Better 
Accuracy 
Higher 
Processing 
Time 
Sadaf Hussain 
Janjua et.al. 
Multi-level aspect-based 
sentiment classification of 
Twitter data: using a hybrid 
approach in deep 
learning[23] 
Artificial Neural 
Network Multilayer 
Perceptron 
80.38 Higher 
Accuracy 
Higher 
Processing 
Time 
Bartosz 
Krawczyk et.al. 
Sentiment classification from 
multi-class imbalanced 
Twitter data using 
binarization[24] 
one-vs-one binary 
decomposition and 
dimension reduction 
66.36 
(G-mean) 
Average 
metrics for 
the 
classification 
Higher 
Processing 
Time 
 
 
The above methods discussed are based on different 
features of the tweet text, such as unigram. Bigram.  
N-gram features, Sentiment features, Punctuation 
features, Syntactic/stylistic features, and semantic 
features are used for the classification. Various feature 
extraction studies show that word embedding based on 
language models is more contextual than the other 
methods. The above studies reveal that the deep learning 
models outperform the baseline classifiers in terms of 
accuracy and processing time, as well as the need for a 
better classifier for effective multi-label classification in 
the disaster scenario. The proposed model works on the 
text as the feature using Crisis word vector-specific   
ELMo embedding with ensemble CNN classifier for the 
multi-classification of the microblog text, which creates 
contextualized feature vector and feature map, resulting 
in improved performance of the classifier.  
 
 
 
3   Proposed Methodology 
 
The proposed EWECNN method was devised with three 
underlying functional modules. Enhanced-ELMo handles 
disaster-related vectors; Enhanced ELMo-CNN 
Architecture extracts context-related features, and Crisis 
Word vector-specific CNN and RNN stacks classify text 
into multiple classes and significantly improve this multi-
label classification. Instead of converting text as word 
vectors and running the algorithm, this model focuses on 
creating context-related, disaster word-specific vectors as 
a  whole sentence rather than single words. This section 
describes the fundamental functional building 
components in further detail. Figure I displays the whole 
flowchart for the EWECNN approach. The algorithm I: 
EWECNN algorithm explains the methodology in a 
stepwise procedure. 

134 Informatica 46 (2022) 131–144                                                                                                                             E. Arathi et al.
  
  
 
 
 
Algorithm I :  EWECNN Algorithm  
                                                                                      
The proposed EWECNN method for multi-label 
classification of tweets with crisisNLP dataset 
Algorithm I : EWECNN(dataset d) 
Begin 
//  input layer- creating word embeddings using context 
based language model . 
 load dataset d 
 create training, validation, and test sets of data 
 For each tweet t in d 
 begin 
     clean_tweet<-preprocess_tweet(t) 
     feature_vector <- enELMo(clean_tweet) 
              end 
// enELMo-CNN layer   
 feature_map = ECA(feature_vector) 
// Multi-label classification with ensemble CNN and 
RNN. 
 z <- CWV_CRNN(feature_map) 
 labels<-output(z) 
end 
 
3.1 Enhanced ELMo to generate crisis word 
vectors  (EECWV) 
This module is the model's input layer, which adds the 
disaster's context to the word vector for categorization,  
then transferred to the following phase. For the 
convolutional net, vectors are generated using 
embeddings. ELMo can build a contextually rich 
representation of words and dynamically alter the 
representation of words, hence resolving polysemous 
words. A deep bidirectional language model (biLM) is 
pre-trained on a large crisis lexicon corpus[32] to 
generate these word vectors. Adding these vectors above 
each input word for each end task significantly improves 
the algorithm's performance. enELMo representations are 
derived from crisis-related word vectors using the 
RLSTM encoder. The ELMo parses a sequence of word 
vectors at a time to eliminate the ambiguous meaning 
which misleads in classification. This context-biased 
smart purport selection is the primary reason to designate 
the enELMo model here. The steps are listed as an 
algorithm II – enELMO algorithm. 
 
Algorithm II: enELMO algorithm  
This algorithm process the feature vector of the given 
dataset using enhanced ELMo 
Algorithm enELMo(preprocess_tweet) 
Begin 
rlstm<-lstm(crisis_word_vector) using equation(4) 
crisis_based_embedding<-ELMo(rlstm) using  
equation(3) 
feature_vector<crisis_based_embedding(preprocess 
                                                                       _tweet) 
return feature_vector 
end   
 
ELMo is often suitable for domain-specific 
categorization. This model achieves domain specificity 
by interacting with the intermediate layer of biLM 
models that are initialized by crisis word vectors. For 
every token t_k, a set of 2L+1 depictions for L-Layer 
using the following equation. 
 
𝑅 𝑘 = {𝑥 𝑘 𝐿𝑀
,ℎ
⃗ 
𝑘 ,𝑗 𝐿𝑀
,ℎ
⃗ 
′
𝑘 ,𝑗 𝐿𝑀
,| 𝑗 = 1 → 𝐿 }              (1) 
ℎ
𝑘 ,𝑗 𝐿𝑀
= {ℎ
⃗ 
𝑘 ,𝑗 𝐿𝑀
;ℎ
⃗ 
′
𝑘 ,𝑗 𝐿𝑀
}                    (2) 
By disavowing the forward and backward biLM iteration 
directions using Equation 2 and substituting it in 
equation 1,  
𝑅 𝑘 = {ℎ
𝑘 ,𝑗 𝐿𝑀
 | 𝑗 = 0 → 𝑁 }                            (2) 
Where ℎ
𝑘 ,0
𝐿𝑀
 is the token layer for each biLSTM layer 
ELMo merged all layers to generate a word vector R as 
𝐸𝐿𝑀𝑜 𝑘 = {𝐸 (𝑅 𝑘 ;𝛩 𝑒 )}. The domain  𝜕 specific weight of 
ELMo is calculated as 
𝐸𝐿𝑀𝑜 𝑘 𝜕 = 𝛿 𝜕 ∑ 𝑠 𝑗 𝜕 𝐿 𝑗 =0
ℎ
𝑘 ,𝑗 𝐿𝑀
                 (3) 
Where 𝑠 𝜕 are the softmax normalized weights and 
𝛿 𝜕 refers to the scalar parameter. The enhanced ELMo 
architecture is given in Figure II. 
 
Figure II: Enhanced ELMo Architecture 
 
By adding a crisis lexicon weight reformer logic to 
the standard LSTM model, EECWV module intercepts 
Figure 1: EWECNN Flow Diagram 

A Multi-label Classification of Disaster-Related Tweets…                                      Informatica 46 (2022) 131-144   135 
are made better. Figure III represents this modified 
LSTM as a Reformed LSTM (RLSTM). For every word 
vector at the disposal position by the existing forget gate, 
the sustainability of those vectors is increased by 
doubling the weight while one of the words exists in the 
crisis lexicon. Let 𝜔 be the set of weights of word 
vectors 𝜔 𝑖 , represented as 𝜔 = {𝜔 0
,𝜔 1
,…𝜔 𝑛 } with the 
condition 0 ≤ 𝜔 𝑖 ≤ 1, where 𝑛 is the maximum number 
of word vectors. The presence of the crisis-related word 
is ensured by the crisis lexicon reformer equation 4. The 
reformed LSTM Architecture is illustrated in Figure iii. 
 
 
                                                             (4) 
 
Figure III: Reformed LSTM 
 
The suggested EECWV module is formulated by 
substituting the RLSTM for every conventional LSTM in 
the ELMo model. This weight retention for vectors 
containing crisis-related terms can endure for longer, 
resulting in an improved context-sensitive model. The 
equations for input, output, and forget gates are given in 
(Eq. 9, 10,11). 
 
3.2  EnELMo CNN Architecture (ECA)  
CNN is often used extensively for image categorization 
[33][34]. Since texts are single dimension character 
arrays sharing the characteristics of time series data,           
1D-kernel is used for the convolution process. 
Generating vectors through word embeddings from 
language models enables the enhanced contextual feature 
extraction through CNN architecture. The steps of this 
module are explained using algorithm III – ECA 
algorithm. The architecture of this module is shown in 
figure IV. 
 
Algorithm III:  ECA(vector vm) 
begin 
 Input the feature vector 
 vmwm<- 1D- Convolutiol_net(vm)  
 Kernel size(k) <- Equation(5) 
 Vmwmkm<- ReLU(vmwm, kernel km) 
 c<-MaxPool (vmwmkm) 
 return(cm) 
end 
 
Figure IV: ECA Architecture 
 
The word to vector conversion by enhanced ELMo 
ensures a more dense representation of context-sensitive 
input texts. There is also a greater probability of getting 
tangled in more than one embedding with similar 
contextual properties in the same direction. Therefore, a 
single dimension convolutional process takes place here. 
The procedure of 1D convolutional operation for 
extracting feature maps is represented in Figure V below. 
 
Figure V: Enhanced ELMo CNN Architecture 
 
In Figure V, v1, v2, v3…vN refers to the input 
vectors supplied by the EECWV module, w1, w2…wM 
refers to the weights assigned to the vectors, k1, k2 … k 
X refers to the kernels, and f1, f2,…,f M refers the 
element values of the kernels. The result of the 
convolutions are represented as c1, c2,…cM. The 
convolution operation is the multiplication and addition 
cross input matrix and filter matrix resulting in a feature 
map(V nW n). All the input elements are covered using a 
sliding window of the filter matrix with a stride of 1[35]. 
The kernel size 𝑆 𝑘𝑒𝑟𝑛𝑒𝑙 influences the accuracy of the 
classification directly [36]. Therefore, in it is calculated 
dynamically in the proposed ECA model by Equation 11, 

136 Informatica 46 (2022) 131–144                                                                                                                             E. Arathi et al.
  
𝑆 𝑘𝑒𝑟𝑛𝑒𝑙 =
max(𝑠𝑖𝑧𝑒𝑜𝑓 (𝑉 1
),𝑠𝑖𝑧𝑒𝑜𝑓 (𝑉 2
),…𝑠𝑖𝑧𝑒𝑜𝑓 (𝑉 𝑁 )) +
𝑠𝑖𝑧𝑒𝑜𝑓 (𝑝𝑎𝑑𝑑𝑖𝑛𝑔 )                                         (5) 
 
where 𝑉 1
,𝑉 2
,…𝑉 𝑁 are the embedding vectors in 
which 𝑁 refers to the number of vectors. The kernel size 
𝑆 𝑘𝑒𝑟𝑛𝑒𝑙 comprehends the values of 𝑋 and 𝑀 . The output 
of this CNN layer is the column vector(c m), which is the 
input text's feature map for further classification. The 
advantages of the above ELMo-CNN model are better 
detection of pertinent patterns from the text, such as 
bigrams, trigrams, and n-grams. The independent 
placement property of CNN makes it possible to detect 
these features from a sentence irrespective of the 
keywords' positions.  
 
3.3 Crisis word vector-specific CNN and 
RNN stacks (CWV-CRNN) 
CNN and RNN stacks are introduced here to acquire 
higher accuracy classification results of the streaming 
temporal text data. The output c1,c2,…cM of the ECA 
module is fed to the input section of the CWV-CRNN. 
The contextual convolutional layer (C-CN)  is the most 
critical layer in the CRNN module. This module 
performs a number of iterations of the convolution over 
the sentence matrix. The C-CN provides a new 
interpretation of the meaning for the words by utilizing 
the crisis word lexicon connected to this module [30]. 
The context region for updating the current position is 
determined by the filter size m and the preset iteration 
step T : (m − 1)T + 1. After going through the non-linear 
activation function and adding the input, the C-CN 
attains non-linear word relationships, enhancing the 
meaning. An output unit at the i
th
 index of the j
th
 feature 
map at layer step t is defined by the equation: (6). 
 
 
𝑓 (𝑤 ) = (𝛤 ∑ ((𝑊 𝑖 ⊗ 𝜔 )+ 𝑏 𝑖 )
N
𝑖 =1
)                         (6) 
 
Where 𝑊 1
,𝑊 2
,…𝑊 𝑁 are vector weight matrices, 
𝑏 1
,𝑏 2
,…𝑏 𝑁 are corresponding biases and ⊗ is the 
elementwise multiplication operator. When the dot 
product of the embedded word vectors is used to detect 
commonalities between them, the interdependence of 
words for a specific class can be determined. 
The CNN architecture used in this phase is given in 
Figure VI. The process of this module is listed below as 
algorithm VI  – CWV-CRNN algorithm. 
 
Algorithm IV -  CWV-CRNN(feature_map) 
Begin 
//ensemble context aware convolutional net. 
 Conv_map <- convolutional_net(feature_map, 
crisis word lexicon) 
 Rectified_map<-relu(Conv_map) 
 conv_map<- convolutional_net(rectified_map) 
 Rectified_map<-relu(conv_map) 
 Reduced_map<-maxpool(rectified_map) 
 Context_map<-dense(reduced_map) 
//classification determinator 
 P(c)<- Equation (6) 
 Classification_index <- Equation(7)  
 if Classification_index < 0.5 
 then 
     classification_index= rnn(context_map) 
 end if 
//classification output 
  z<-label; 
  
 
 
  
 
Figure VI: Architecture of  CWV-CRNN Module 
 
The sigmoid activation function is utilised at the dense 
layer of the model with the probability of a class (c t) as 
Bernoulli distribution (Eq. 15 , 16). Now the probabilities 
of each class are independent of the other class 
probabilities. Based on the activation function, the 
threshold value is set to 0.5. At this stage, the output of 
this section(z) is highly accurate, and the CWV-CRNN 
module skips the subsequent RNN stack to conserve 
processing time. (Equation. 7) is used to decide whether 
the results are pertaining to the accuracy or not. 
 
{
ℎ𝑖𝑔 ℎ𝑙𝑦 𝑎𝑐𝑐𝑢𝑟𝑎𝑡𝑒 𝑖𝑓 max(𝛤 1
,𝛤 2
…𝛤 8
) ≥ 2𝜎 𝛤 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑜𝑡 ℎ𝑒𝑟𝑤𝑖𝑠𝑒                                          
     (7)                             
Where 𝛤 1
→ 𝛤 8
 are the acquired classification index 
of desired classifications, 𝜎 𝛤 indicates the standard 
deviation of every classification index value. If the 
output is less than the desired index, it is passed to the 
RNN module to improvise the classification index. RNN 
has obtained much attention because of its ability to 
preserve the sequence information over time. The 
embedding of the target tweet content and the contextual 
features are concatenated before sending to the LSTM 
z = 

A Multi-label Classification of Disaster-Related Tweets…                                      Informatica 46 (2022) 131-144   137 
layer. This generates compositions for entire sentences 
using RNN  for the content and context data, which helps 
the sentences to a higher dimensional space and an 
output as numerical values for the context based on the 
content[37]. In this paper, RNN is utilized to capture past 
and future information. So, the output of the i-th word is 
shown in the following equation (8). 
 
   ℎ𝑖 = ℎ
⃗ 
𝑖 ⊕ h
⃗ 
𝑖                                                       (8) 
Here, ⊕ denotes the concatenation of hidden states 
of the forward LSTM and the backward LSTM. The 
RNN stack equations for input, output, and forget gates 
of CWV-CRNN module using the equations (9), (10), 
(11), (12), (13), and (14). 
𝑖 𝑡 = 𝛤 (𝑊 𝑥𝑖
𝑥 𝑡 + 𝑊 ℎ𝑖 ℎ
𝑡 −1
+ 𝑊 𝑐𝑖
𝑐 𝑡 −1
+ 𝑏 𝑖 )                (9) 
𝑓 𝑡 = 𝛤 (𝑊 𝑥𝑓
𝑥 𝑡 + 𝑊 ℎ𝑓 ℎ
𝑡 −1
+ 𝑊 𝑐𝑓
𝑐 𝑡 −1
+ 𝑏 𝑓 )              (10) 
𝑜 = 𝛤 (𝑊 𝑥𝑜
𝑥 𝑡 + 𝑊 ℎ𝑜 ℎ
𝑡 −1
+ 𝑊 𝑐𝑜
𝑐 𝑡 −1
+ 𝑏 𝑜 )                (11) 
𝑐 𝑡 = 𝑓 𝑖 𝑐 𝑡 −1
+ 𝑖 𝑡 𝑅𝑒𝐿𝑈 (𝑊 𝑥𝑐
𝑥 𝑡 + 𝑊 ℎ𝑐 ℎ
𝑡 −1
+ 𝑏 𝑐 )          (12) 
ℎ
𝑡 = 𝑜 𝑡 𝑅𝑒𝐿𝑈 (𝑐 𝑡 )                 (13) 
𝑦 𝑡 = 𝑤 𝑐 𝑦 𝑐 
𝑡 + 𝑤 𝑐 `
⃗⃗ ⃗ 
𝑦 𝑐 `
⃗⃗ 
𝑡 + 𝑏 𝑦               (14) 
The CWV-CRNN module receives 𝑁 number of 
input vectors and produces 8 classification index values 
by which the classifications are determined either by 
positive or negative by comparing with the threshold 
value 𝜏 . Classification type notations and their equivalent 
classification labels are described in Table ii. The 
classification determinator equation is given below.  
 
𝑃 (
𝑐 𝑡 𝑥 𝑖 ) = 
1
1+exp (−𝑧 𝑡 )
                       (15) 
∀𝑖 = 1 → 7 ≔ 𝛤 𝑟 𝑖 = {
𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑖𝑓 𝛤 𝑖 ≥ 𝜏   
𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑜𝑡 ℎ𝑒𝑟𝑤𝑖𝑠𝑒              (16) 
Table II: Representations of  Labels in CRNN Module 
       Label 
 𝜞 𝟏 Disaster type 
 𝜞 𝟐 Location 
 𝜞 𝟑 Dead and Injured People 
 𝜞 𝟒 Help Request 
 𝜞 𝟓 Infrastructure damage 
 𝜞 𝟔 Search and Rescue 
 𝚪 𝟕 Weather-related 
 𝚪 𝟖 Non-relevant information 
 
 
4   Experimental Setup 
A computer powered by Intel Core i7, 3770K 3.5GHz 
Quad-Core processor with 16GB RAM and 1TB storage, 
is used to develop and test the proposed method. A 
dedicated User Interface (UI) is designed using Visual 
Studio Integrated Development Environment [37]. The 
UI is responsible for loading the dataset and evaluating 
the proposed method's performance. Visual C++ [38] 
programming language is used to code the functional 
modules of the proposed method. Existing and proposed 
methods are evaluated sequentially, and the performance 
metrics are logged into corresponding memory 
references. The UI is also designed to fetch the 
evaluation metric results from memory and export them 
as a report file. Graphs are generated in a separate  
window for all classifications. The UI can get the user's 
dynamic Twitter posts and classify them. CrisisNLP 
Twitter dataset [39] is used to measure the classification 
performance of the EWECNN method. Since the 
proposed method is context-sensitive, the preprocessing 
for input has only a negligible impact on the results. 
Therefore, the basic preprocessing of removing URLs, 
links, similes, and emoticons is done in this work to 
preserve processing complexity. Dataset is randomly 
divided into ten equal parts for the further process. 
4.1 Dataset 
We utilise data from the CrisisNLP standard dataset of 
tweets [35, 36]. The sources are tweets sent during 
several natural disasters and annotated by paid 
employees and volunteers. The dataset includes, among 
other occurrences, earthquakes, floods, and typhoons.. 
Tweets are categorized into many informative categories 
such as urgent needs, contribution offers, infrastructure 
damage, dead or injured persons, and irrelevant or 
unrelated class. A single-line description for each 
category is in Table iii. 
 
Table III: Description of the classes in the datasets 
Class  Description 
Disaster_type Disaster event  
Location Disaster location 
Dead_and_Injured  Reports of deaths, and injuries, of the people during disaster 
Help_Request      Messages containing donations (food, shelter, services, etc.) or 
volunteering offers 
Infrastructure_damage Reports of infrastructure and utility damage 
Search_and_Rescue Messages of search operations and rescue operations during crisis 

138 Informatica 46 (2022) 131–144                                                                                                                             E. Arathi et al.
  
Weather_related Messages containing helpful information about the disaster event 
that does not fit in one of the above classes 
irrelevant  Irrelevant or not informative, or not beneficial for crisis response 
 
 
4.2 Classification 
Dataset is split into 80% of the dataset as the training set 
and 20% as the test set. The validation set is made up of 
20% of the training set. The word vectors were initialized 
using pre-trained crisis word embeddings using 
EMTerms. Multi-classification of Twitter texts is 
conducted using the Enhanced – Elmo – CNN feature 
extraction approach and ensemble stack of CRNN. The 
classification estimator of the ensemble stack improves 
the model's performance by enhancing the accuracy of 
label prediction and shortening the processing time. The 
entire dataset is divided into 10 equal parts, and the 
values for every 10 percent of the data are input into the 
model and recorded in the report. 
 
5   Results and Discussion 
There are 8 customarily used classifications as Disaster 
type, location, Dead and Injured people, Help request,  
Infrastructure damage,  Search and Rescue,  Weather-
related and non-relevant information are targeted in this 
work. The Twitter NLP Crisis dataset is classified, and 
its classification performance parameters such as 
Accuracy, Precision, F-Score, and G-Mean [39] are 
discussed in this section. 
5.1 Performance metrics of the proposed 
EWECNN Model with related models. 
  The metrics of the proposed model for multi-label 
classifications of the microblog text for the disaster 
events rescue are depicted with plots and discussed in 
detail in this section. Overall performance metrics scores 
of the proposed method and the existing methods 
discussed in section 2 are represented using table iv. 
 
Table IV: Performance metrics score of the proposed 
EWECNN Model with the other methods. 
 
Methods Accuracy Precision F1-
Score 
G-
Mean 
MC_BDACSA 81.41 81.29 81.39 81.39 
MC_PAMSA 79.76 79.79 79.76 79.76 
MC_CRICCD 88.05 87.87 88.03 88.03 
MC_MASC 83.00 83.13 83.02 83.02 
MC_BMSC 85.58 85.28 85.54 85.54 
MC_EWECNN 90.46 90.43 90.46 90.46 
 
The above table – iv clearly shows the proposed model – 
EWECNN has higher accuracy than other methods 
discussed in this work. The proposed EWECNN method 
uses context-specific feature mapping and context-based 
classification using the EMTerms lexicon, which has 
improved the multi-label categorization of the tweets 
more than the other methods discussed here. In the other 
methods, the CRICCD method, which uses word 
embedding for feature vectorization, has better accuracy 
than the PAMSA, MASC, BMSC, and BDACSA 
methods discussed in section 2. 
5.1.1  Accuracy  
Accuracy refers to the classification results' proximity to 
the actual classified labels. The formula for calculating 
accuracy is (TP+TN)/(TP+TN+FP+FN) [40]. Figure VII 
shows a thorough Accuracy log comparison graph of the 
proposed model with existing approaches. Table iv 
shows the calculated accuracy values for classifications 
performed by existing methods and the proposed model. 
As per the observed results, the proposed EWECNN 
method achieved the highest accuracy average of 
93.49%, which is 12.4% higher than the successive 
performer CRICCD. The ranking of the  classification 
methods based on the accuracy are EWECNN, CRICCD, 
BMSC, MASC, BDACSA, and PAMSA with 93.5% 
89.0 %, 86.6 %, 85.1 %, 83.0 %, and 81.10 % 
respectively.    
 
 
Figure VII Accuracy log comparison graph of                     
the proposed EWECNN model with existing approaches 
 
5.1.2      Precision 
Precision is directly proportional to a classification 
algorithm's quality. A higher precision value refers 
higher quality of the algorithm. Precision is calculated 
using the formula TP/(TP+FP). A comparison of the 
proposed EWECNN's precision scores with the existing 
methods is represented in figure VIII. The score is listed 
in table iii. Based on the observation results, the  
performance rank sequence for proposed and existing 
methods in terms of precision is EWECNN, CRICCD, 
BMSC, MASC, BDACSA, and PAMSA, with values of 
93.47 % ,89.55 %, 86.33 %, 83.76 %, 81.93 %, and 

A Multi-label Classification of Disaster-Related Tweets…                                      Informatica 46 (2022) 131-144   139 
80.42 % respectively. The proposed EWECNN approach 
achieves the greatest precision of 93.47%, which is 3.9 
percent greater than the closest precision score of 
CRICCD. During the trial, it was discovered that the 
proposed EWECNN approach outperformed the other 
methods regarding the classification-wise precision 
average. 
 
 
Figure VIII  Comparison plot of the proposed EWECNN 
model with existing approaches for the performance 
metrics - Precision 
5.1.3 F1-Score 
The quality of a classification algorithm is directly 
proportional to the F-Score index. F-Score is calculated 
using the equation TP/(TP+1/2(FP+FN)). Table iii 
compares the F1-Score of the proposed model with the 
existing approaches.Classification wise F-Score averages 
of EWECNN, CRICCD, BMSC, MASC, BDASCA and 
PAMSA are 92.99%, 89.84%, 86.61%, 85.12%, 83.00% 
and 81.11% respectively. The proposed EWECNN 
approach achieves the greatest F1-Score of 93.49%, 
which is 3.15% greater than the closest precision score of 
CRICCD. Figure IX shows a comparative graph of the F-
Score values of the proposed method with other methods 
discussed in this work. 
 
 
Figure IX: A Comparative graph of the F1-Score metric 
of the proposed EWECNN method with other methods 
5.1.4 Geometric Mean(G-Mean) 
The G-Mean is a statistic metric that evaluates the 
proportional performance of majority and minority 
classifications. Even if negative instances are correctly 
categorized, a low G-Mean suggests poor performance in 
classifying positive cases. This value is essential for 
preventing overfitting the negative class and underfitting 
the positive class. G-Mean is calculated using the 
formula √(TP/(TP+FP)×TP/(TP+FN)). Table iii depicts 
the G-Mean of the proposed method and existing 
methods. The highest GM value of 93.29% is achieved 
by the EWECNN method, which is 3.44% higher than 
the nearest achiever, CRICCD. Performance ranking 
sequence in terms of G-Mean is EWECNN, CRICCD, 
BMSC, MASC, BDACSA, and PAMSA, with the values 
93.29%, 89.85%, 86.61%, 85.12%, 83.00%, and 81.11% 
given in order from the best score of G-Mean. The 
comparison graph is given below in Figure X. 
 
 
Figure X: Comparison of  G-Mean metrics of the 
proposed EWECNN model with other models 
 
 
Based on the observed result, it is understood that the 
proposed EWECNN method secures higher performance 
scores in terms of Accuracy, Precision, Sensitivity, 
Specificity, F1-Score, and G-Mean. 
 
5.5 Performance analysis of the multi-label 
classification of the tweets using the 
proposed model. 
The proposed EWECNN model multi-classifies the tweet 
text under eight categories. The suggested model's 
cumulative report analysis for all eight classes is 
discussed in this section. Table v shows the actual labels 
as well as the categorized labels 
 
Table V: Count of Actual labels and Predicted labels 
Class  #Actual labels #Predicted  
Labels 
Disaster type 6,418  5969 
Location 5365 4882 
Dead and Injured 2,437 2266 
Help Request      3,683 3425 
Infrastructure damage 3,288  3058 
Search and Rescue 6,178 5746 
Weather-related 10,059 9119 

140 Informatica 46 (2022) 131–144                                                                                                                             E. Arathi et al.
  
Irrelevant  12137  13300 
 
Table vi shows the classification efficiency of the 
labels using the proposed method with the performance 
metrics – Accuracy, Precision, F-Score, and G-Mean. 
The tweets are assigned with these labels during a 
disaster event by the proposed EWECNN model. These 
labels aid in faster rescue operations. 
 
 
Table VI: Performance metrics of classification of the 
labels by the proposed EWECNN Model. 
 
S.No Classification 
Labels  
Accuracy Precision F1-
Score 
G-
Mean 
1 Disaster type 91.82 91.42 91.57 92.03 
2 Location 92.29 91.78 92.26 92.26 
3 Dead and 
Injured 93.39 93.37 93.39 93.39 
4 Help Request      91.16 92.11 91.34 91.66 
5 Infrastructure 
damage 93.94 93.12 92.95 92.95 
6 Search and 
Rescue 93.08 92.96 93.07 93.07 
7 Weather-
related 94.16 93.53 94.16 94.14 
8 Irrelevant 93.86 93.82 93.75 93.80 
 
The above depicts the categorization of the tweet texts 
based on the labels. Among the eight labels, the 
categorization of help requests has the maximum 
accuracy than other labels. and has the least accuracy 
score among the labels. The overall accuracy average of 
the multi-classification of the label is 93.3% which is 
higher than the nearest model CRICCD. Figure XI 
presents the comparison graph on the performance 
metrics of the proposed EWECNN model. 
 
 
 
Figure XI: Performance analysis of the multi-label 
classification of the tweets using the proposed model 
 
According to the plot presented above, the label 
requesting assistance has higher metrics than the other 
labels. When classifying the type of disaster, 91.82 
percent accuracy was achieved. The proposed method  
achieves  superior performance than the other method 
that has been discussed. These models employs  baseline 
classifiers, perceptron classifiers, and deep learning 
classifiers for the classification of tweets . 
This study demonstrates that using context-specific 
ELMo in conjunction with an ensemble of 1D-CNNs and 
RNNs is an effective way to improve multi-label 
classification and reduce processing time.                 
 
6   Conclusion 
 
At present, the contribution of social media during any 
crisis is notably comparable with conventional media. At 
the same time, the ratio of irrelevant content in social 
media is much higher than in other conventional media. 
By incorporating a suitable classification procedure to 
deal with social media data, the impact of the irrelevant 
data can be reduced meaningfully. In this work, an 
improved classifier is introduced in the name 'EWECNN' 
to parse the Twitter data meticulously and classify them 
into given categories. The introduction of a crisis lexicon 
influenced LSTM into ELMo, Context-sensitive variable 
kernel CNN, and competent classification estimator 
based CRNN Stack are the novelties that contributed to 
this work. Though several existing methodologies are 
available to perform the multi-classification task, the 
novelties introduced in the proposed modules such as 
EECWV, ECA, and CRV-CRNN ensure higher 
classification accuracy. As per the experiments carried 
out with the benchmark dataset, it is asserted that the 
performance of the proposed EWECNN method has 
surpassed other methods. Due to this work's novelty and 
performance improvements, EWECNN can be used 
lively to support society during the crisis in a better way. 
 
7   References 
[1] [1] Olteanu A, Vieweg S and Castillo C. What 
to expect when the unexpected happens: Social 
media communications across crises.,   
Proceedings of the 18th ACM conference on 
computer-supported cooperative work & social 
computing.ACM;.994–1009,2015.          
 
[2] Gralla E, Goentzel J and Van de Walle B. 
Understanding the information needs of field-
based decision-makers in humanitarian response 
to sudden-onset disasters.,  ISCRAM; 2015. 
 
[3] Y. Agarwal, D. K. Sharma, and R. Katarya, 
Sentiment/Opinion Review Analysis: Detecting 
Spams from the good ones!, 2019, 4th 
International Conference on Information 
Systems and Computer Networks (ISCON),  

A Multi-label Classification of Disaster-Related Tweets…                                      Informatica 46 (2022) 131-144   141 
557-563,2019. 
http://doi.org/10.1109/ISCON47742.2019.9036
249. 
 
[4] Imon Banerjee, Yuan Ling, Matthew C. Chen, 
Sadid A. Hasan, Curtis P. Langlotz, Nathaniel 
Moradzadeh, Brian Chapman, Timothy 
Amrhein, David Mong, Daniel L. Rubin, 
Oladimeji Farri, Matthew and P. Lungren, 
Comparative effectiveness of convolutional 
neural network (CNN) and recurrent neural 
network (RNN) architectures for radiology text 
report classification, Artificial Intelligence in 
Medicine,  97,  79-88, 2019.  
https://doi.org/10.1016/j.artmed.2018.11.004  
 
[5] Manzhu Yu, Qunying Huang, Han Qin, Chris 
Scheele and Chaowei Yang, Deep learning for 
real-time social media text classification for 
situation awareness – using Hurricanes Sandy, 
Harvey, and Irma as case studies, in 
International Journal of Digital Earth,    
1230-1247, 2019. 
http://doi.org/10.1080/17538947.2019.1574316. 
 
[6] M. P. Akhter, Z. Jiangbin, I. R. Naqvi, M. 
Abdelmajeed, A. Mehmood, and M. T. Sadiq, 
Document-Level Text Classification Using 
Single-Layer Multisize Filters Convolutional 
Neural Network, IEEE , 8,       
42689 - 42707, 2020, 
http://doi.org/10.1109/ACCESS.2020.2976744. 
 
[7] H. Ma, Y. Li, X. Ji, J. Han, and Z. Li,  MsCoa: 
Multi-Step Co-Attention Model for Multi-Label 
Classification, IEEE Access, vol. 7, pp. 109635-
109645, 2019,  
http://doi.org/10.1109/ACCESS.2019.2933042.  
 
[8] Alsaedi, N., Burnap, P. and Rana, O.  Can We 
Predict a Riot? Disruptive Event Detection 
Using Twitter, ACM Transactions on Internet 
Technology, 17(2), 18, 2017. 
https://doi.org/10.1145/2996183.  
 
[9] Starbird, K., Palen, L., Hughes A.L., and 
Vieweg, S. Chatter on the red: what hazards 
threat reveals about the social life of 
microblogged information. Proceedings of 
CSCW. ACM, 241–250, 2010. 
        https://doi.org/10.1145/1718918.1718965. 
 
[10] Vieweg, S., Castillo, C. and Imran, M. 
Integrating social media communications into 
the rapid assessment of sudden onset disasters. 
In Social Informatics, Springer, 444–461, 2014. 
       https://doi.org/10.1007/978-3-319-13734-6_32 
 
[11] Verma, S., Vieweg, S., Corvey, W., Palen, L., 
Martin, J., Palmer, M., Schram, A., and 
Anderson, K. , Natural Language Processing to 
the Rescue? Extracting "Situational Awareness" 
Tweets During Mass Emergency. Proceedings 
of the International AAAI Conference on Web 
and Social Media, 5(1), 385-392., 2021. 
Retrieved from 
https://ojs.aaai.org/index.php/ICWSM/article/vi
ew/14119.  
 
[12] Imran, M., Castillo, C., Lucas, J., Meier, P., and 
Vieweg, S., Aidr: Artificial intelligence for 
disaster response, in Proceedings of the 23rd 
International Conference on World Wide Web, 
ACM. pp. 159–162, 2014. 
        https://doi.org/10.1145/2567948.2577034. 
 
[13] Sreenivasulu and M., Sridevi, M., Mining 
informative words from the tweets for detecting 
the resources during disaster, International 
Conference on Mining Intelligence and 
Knowledge Exploration, Springer, 348–358, 
2017. 
       https://doi.org/10.1007/978-3-319-71928-3_33. 
 
[14] Rudra, K., Ganguly, N., Goyal, P.,and Ghosh, 
S.,. Extracting and summarizing situational 
information from Twitter social media during 
disasters. ACM Transactions on the Web 
(TWEB) 12(17), 2018. 
        https://doi.org/10.1145/3178541. 
 
[15] Rudra, K., Ghosh, S., Ganguly, N., Goyal, P., 
and Ghosh, S.,  Extracting situational 
information from microblogs during disaster 
events: a classification-summarization approach, 
in Proceedings of the 24th ACM International 
Conference on Information and Knowledge 
Management,  583–599, 2015. 
        https://doi.org/10.1145/2806416.2806485. 
 
[16] Caragea, C., Silvescu, A. and Tapia, A.H.,  
Identifying informative messages in disaster 
events using convolutional neural networks, in 
International Conference on Information 
Systems for Crisis Response and Management., 
2016. 
 
[17] Nguyen, T.D., Al-Mannai, K., Joty, S.R., Sajjad, 
H., Imran, M., and Mitra, P. , Rapid 
Classification of Crisis-Related Data on Social 
Networks using Convolutional Neural 
Networks. ArXiv, abs/1608.03902, 2016. 
 
[18] Al-Rasheed, Amal A. “Finding Influential Users 
in Social Networking Using Sentiment 
Analysis.” Informatica, 46(5), Mar. 2022.  
               https://doi.org/10.31449/inf.v46i5.3829.  
   
        [19] Etaiwi, Wael & Suleiman, Dima and Awajan,    
               Arafat,. Deep Learning Based Techniques for  
               Sentiment Analysis: A Survey. Informatica.  
               45. 89-96. ,2021. 

142 Informatica 46 (2022) 131–144                                                                                                                             E. Arathi et al.
  
       https://doi.org/10.31449/inf.v45i7.3674.  
 
   [20]  Gaafar, Alaa Sahl, Jasim Mohammed Dahr,     
                 and Alaa Khalaf Hamoud.,Comparative  
                 Analysis of Performance of Deep Learning  
                 Classification Approach Based on LSTM- 
                 RNN for Textual and Image Datasets., 
                 Informatica 46(5), 2022.           
                 https://doi.org/10.31449/inf.v46i5.3872. 
 
        [21]  Nguyen, D.T., Al Mannai, K.A., Joty, S.,      
   Sajjad, H., Imran, M.and Mitra, P., Robust  
   classification of crisis-related data on    
   social networks using convolutional  
   neural networks, Eleventh nternational  
   AAAI Conference on Web and Social  
                 Media., 2017. 
  
        [22] Mikolov, T., Sutskever, I., Chen, K.,  
  Corrado, G.S.  and Dean, J.,  Distributed  
  representations of words   and phrases and  
  their compositionality, Advances in neural  
  information processing systems, 3111– 
  3119, 2013. 
 
        [23] Madichetty, S. and Sridevi, M., Detecting  
                informative tweets during disaster using  
                deep neural networks,, 11th International  
                Conference on Communication Systems &  
                Networks . OMSNETS), 709–713, 2019. 
 
       [24] Rodrigues, A.P.and Chiplunkar, N.N., A new     
               big data approach for topic classification and     
               sentiment analysis of Twitter data", Evol. Intel.  
               2019. 
               https://doi.org/10.1007/s12065-019-00236-3. 
 
       [25] M. Bouazizi and T. Ohtsuki, A Pattern-Based  
               Approach for Multi-Class Sentiment Analysis     
               in  Twitter, in IEEE Access, (5), 20617-20639,   
 2017.               
 http://doi.org/10.1109/ACCESS.2017.2740982. 
 
[26]  Sreenivasulu, Madichetty and Sridevi,     
Improved Classification of Crisis-Related Data    
on Twitter using Contextual Representations,        
Procedia Computer Science, 167,962-968,   
                2020. 
                https://doi.org/10.1016/j.procs.2020.03.395. 
 
        [27]Janjua, S. H., Siddiqui, G. F., Sindhu, M. A.,                                
and Rashid, U. Multi-level aspect based  
sentiment  classification of Twitter data: using 
hybrid approach  in deep learning. PeerJ. 
Computer science, 7, e433. ,         2021.  
               https://doi.org/10.7717/peerj-cs.433. 
 
[28] Krawczyk B., McInnes B.T. and Cano A.  
Sentiment Classification from Multi-class 
Imbalanced Twitter Data Using Binarization. In: 
Martínez de Pisón F., Urraca R., Quintián H., 
Corchado E. (eds) Hybrid Artificial Intelligent 
Systems. HAIS 2017. Lecture Notes in 
Computer Science, vol 10334. Springer, Cham., 
2017. 
https://doi.org/10.1007/978-3-319-59650-1_3 
 
[29] Olteanu, C. Castillo, F. Diaz, and S. Vieweg, 
Crisislex: A lexicon for collecting and filtering 
microblogged communications in crises, 
Proceedings of the 8th International AAAI 
Conference on Weblogs and Social Media 
(ICWSM" 14), no. EPFL-CONF-203561, 2014. 
 
[30] L.Mou and X. X. Zhu, Learning to Pay 
Attention on Spectral Domain: A Spectral 
Attention Module-Based Convolutional 
Network for Hyperspectral Image 
Classification, IEEE Transactions on 
Geoscience and Remote Sensing,  58(1),     
110-122, 2020,. 
http://doi.org/10.1109/TGRS.2019.2933609. 
 
[31] Y. Sun, B. Xue, M. Zhang, G. G. Yen, and J. 
Lv, Automatically Designing CNN 
Architectures Using the Genetic Algorithm for 
Image Classification, IEEE Transactions on  
Cybernetics, 50(9) . 3840-3854, 2020. 
http://doi.org/10.1109/TCYB.2020.2983860. 
 
[32] Cui, Renhao & Agrawal and Gagan & Ramnath, 
Rajiv. Tweets can tell: activity recognition using 
hybrid gated recurrent neural networks., Social 
Network Analysis and Mining.,2020. 
https://doi.org/10.1007/s13278-020-0628-0. 
 
[33] D. Hu and B. Krishnamachari, Fast and 
Accurate Streaming CNN Inference via 
Communication Compression on the Edge, 
IEEE/ACM Fifth International Conference on 
Internet-of-Things Design and Implementation  
(IoTDI), 157-163, 2020. 
http://doi.org/10.1109/IoTDI49375.2020.00023. 
 
[34] Shin, Joongbo & Kim, Yanghoon & Yoon, 
Seunghyun & Jung and Kyomin. Contextual-
CNN: A Novel Architecture Capturing Unified 
Meaning for Sentence Classification. 2018 IEEE 
International Conference on Big Data and Smart 
Computing,491-494, 018., 2018. 
https://doi.org/10.1109/bigcomp.2018.00079. 
 
[35] Lan, Yangyang & Hao, Yazhou & Xia, Kui & 
Qian, Buyue and Li, Chen,  Stacked Residual 
Recurrent Neural Networks With Cross-Layer 
Attention for Text Classification. 
IEEE Access. PP. 1-1.2020. 
https://doi.org/10.1109/access.2020.2987101. 
 
[36] https://visualstudio.microsoft.com/ 
 

A Multi-label Classification of Disaster-Related Tweets…                                      Informatica 46 (2022) 131-144   143 
[37] https://www.udacity.com/blog/2020/02/microso
ft-visual-c-review.html. 
 
[38] https://crisisnlp.qcri.org/ 
 
[39] HaCohen-Kerner Y, Miller D, Yigal Y.  The 
influence of preprocessing on text classification 
using a bag-of-words representation. PLOS 
ONE 5(5):e0232525,. 2020. 
DOI 10.1371/journal.pone.0232525. 
 
[40] José-Ramón Cano, Pedro Antonio Gutiérrez, 
Bartosz Krawczyk, Michał Woźniak, Salvador 
García, "Monotonic classification: An overview 
on algorithms," in performance measures and 
data sets, Neurocomputing, 341,  168-182,  
2019. 
https://doi.org/10.1016/j.neucom.2019.02.024.-
182, ISSN 0925. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

144 Informatica 46 (2022) 131–144                                                                                                                             E. Arathi et al.