International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023

ISSN: 2395-3470
[Link]

Proposed Model for Text Classification Using Natural

Language Processing Techniques.
1
P Atef Ghalwash,2 Esam Abu Elaseam & 3Omnia Mamdouh
P P P P P

1
PFaculty of computers and artificial intelligence, Helwan university
P

2,3
P P Faculty of commerce and business administration Helwan university, Cairo, Egypt

Abstract
The pathology reports contain unorganized text, and it can be difficult to access or search for
clinical information within them. Automated text classification is an important method that can
help manage and process a large number of digital documents, which are continuously increasing,
by categorizing them into predefined classes. This process plays an important role in extracting
and summarizing information, retrieving text, and answering questions. This research focused on
using the LSTM, CNN, and Logistic Regression algorithms to classify pathology reports. The
dataset used contained research papers with more than 6 pages, specifically cancer documents
categorized into Thyroid, Colon, and Lung cancers. The results showed that the model had
acceptable accuracy levels, with the LSTM algorithm achieving a high accuracy rate of 99%.

Keywords: Text Classification; Natural Language Processing; Long short-term memory;

Convolutional Neural Network.

1 Introduction
In recent years, there has been a significant increase in the amount of digital text documents
available. As a result, it has become crucial to efficiently categorize and organize these
documents. Text classification refers to the process of grouping textual information based on its
content [1]. The primary objective of text classification is to divide unstructured documents into
respective categories based on their content [2]. This process finds application in various fields
such as text summarization, information extraction, information retrieval, question answering, and
sentiment analysis due to the widespread availability of textual information [1].
Text classification is used in different fields such as spam filtering, email routing, topic tracking,
sentiment analysis, and web page classification. However, medical specialization is considered
one of the most crucial areas where text classification plays an essential role.
Pathologists use unstructured and semi-structured pathology reports to record detailed
observations of cells, organs, and tissue specimens. These reports contain a vast amount of
information that is essential for advancing cancer research in areas such as treatment selection,
case identification, prognostication, surveillance, clinical trial screening, risk stratification,
retrospective study, and many others [3].
Retrieving crucial descriptive observations from pathology reports presents a significant challenge
because a considerable part of the diagnosis is written in an unstructured, free-text format. State or
national cancer registries responsible for tracking thousands or even millions of patients must rely

51
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

on human experts to decode and convert this relevant information into a structured and
normalized form. This process of manually coding information is expensive, time-consuming,
prone to errors, and places inherent restrictions on the amount and variety of data that can be
extracted [4].
The cancer research communities have recently turned their attention towards NLP as a solution
to overcome the constraints of manual information extraction from pathology reports [5]. This
study aims to create a Text classification model that uses Logistic Regression and LSTM
techniques to automate the IE process on pathology reports. The primary objective is to develop
an efficient and automated approach to extract information from these reports.
The paper is organized as follows: Section 2 presents a brief review of previous research on text
classification and offers examples of text classifiers that have been developed using various
techniques. In Section 3, the theoretical concepts and principles underlying the proposed text
classifier are discussed in detail. Section 4 gives a detailed description of how the classifier works,
including its pseudocode and algorithms for implementation. Time complexity of the classifier is
examined, and results obtained using the classifier are analyzed in Section 5. Finally, Section 6
concludes the paper with a summary of the work and discusses potential future directions for
research.
.

2 Literature review
The objective of this research is to investigate how well different deep learning algorithms
perform in classifying medical notes with imbalanced disease classes. The study utilized seven AI
models, including a CNN, Transformer encoder, and pre-trained BERT, as well as four traditional
sequence neural network models - RNN, GRU, LSTM, and Bi-LSTM. These models were used to
determine the presence or absence of 16 specific diseases based on discharge summary notes from
patients. The best performing model was MLearn-ATC, achieving an accuracy of 0.93. However,
it should be noted that this technique was not evaluated using a larger dataset or non-binary
classification problems [6].
This paper explores the efficacy of a semi-supervised approach for Telugu news articles. The
study uses semi-supervised clustering for pattern classification after initial categorization. Its goal
is to evaluate the impact of n-gram feature selection on text classification for news articles using
semi-supervised learning techniques. Results indicate that these approaches significantly boost
performance, with Support Vector Machine achieving a 93.64% classification rate [7].
The authors suggest using pre-trained language models and logic rules to create prompts with sub-
prompts for multi-class text classification. This method, called PTR, allows for encoding prior
knowledge of each class into the prompt tuning. The experiments conducted on relation
classification, a complex multi-class classification task, demonstrate that PTR outperforms
existing baselines consistently. These results indicate that PTR has promise as an approach that
utilizes both human prior knowledge and PLMs for challenging classification tasks, achieving an
F1 score of 91.9 [8].

52
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

In this study, an intelligent approach that utilizes natural language processing technology
(NLP) is proposed. The approach consists of three stages. Firstly, a Convolution Neural Network
(CNN) based text classification model is developed to classify construction on-site reports by
analyzing and extracting the features of report text. In the second stage, an improved frequency-
inverse document frequency (TF-IDF) method is employed to analyze the classified construction
report texts. The proposed CNN-based text classification model successfully classifies sentences
describing construction on-site situations into six categories with an overall accuracy of 91.24%.
When compared to other models, the CNN-based text classification model consistently performs
with high accuracy [9].
This research proposes a feature selection method that uses the term frequency distribution
measure. The Naive Bayes and SVM classifiers are employed with two benchmark datasets
(WebKB and BBC). The experimental results demonstrate that the proposed feature selection
method achieves higher classification accuracy. Specifically, the TFDM feature selection method
reaches an accuracy of 96.7% using the NB classifier and 95.1% using the SVM classifier [10].
This paper describes the implementation of Support Vector Machines (SVM) for classifying
English text and documents. Two analytical experiments were conducted to evaluate the selected
classifiers using English documents. The experimental results, based on a dataset of 1,033 text
documents, indicate that the Rocchio classifier performs best when the feature set is small, while
SVM outperforms other classifiers. The analysis shows that the classification rate exceeds 90%
when using more than 4,000 features [11].
This study employed the Doc2vec word embedding method for text classification on two
datasets: the Turkish Text Classification 3600 (TTC-3600) consisting of Turkish news texts and
the BBC-News dataset consisting of English news texts. Deep learning-based CNN and traditional
machine learning classification methods, including Gauss Naive Bayes (GNB), Random Forest
(RF), Naive Bayes (NB), and Support Vector Machine (SVM), were used as classification
methods. The proposed model achieved the highest accuracy of 94.17% on the Turkish dataset
and 96.41% on the English dataset using CNN classification [12].
This paper explores the potential and efficacy of Machine Learning models for text
classification. Traditional machine learning methods, such as Support Vector Machines, Naïve
Bayes, and Random Forests, are examined. The paper emphasizes the importance of other steps in
the text classification process, including preprocessing, lemmatization, and n-gram usage, by
discussing relevant datasets. The results show that the Support Vector Machine model achieved
the highest accuracy rate at 95.13%, while the Random Forest algorithm using 3-grams showed
the lowest accuracy [13].

This study utilizes text mining techniques and combines two supervised machine learning
algorithms, Naïve Bayes and Support Vector Machines (SVM), to create a hybrid model for text
classification. The hybrid model was developed using WEKA tools and Java programming
language. The results indicate that the hybrid model achieved an accuracy of 96.76%,
outperforming both the Naïve Bayes and SVM models, which achieved accuracies of 61.45% and

53
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

69.21%, respectively [14].

Janani introduced a new algorithm, the Optimization Technique for Feature Selection (OTFS),
for text classification and compared it with commonly used classification techniques such as
probabilistic neural network, support vector machine, K-nearest neighbor, and Naïve Bayes. The
results show that the proposed algorithm achieves better accuracy in optimizing features and
classifying text documents based on their content [2].

This paper investigates the use of capsule networks for text classification through an empirical
exploration. While previous research has demonstrated the effectiveness of capsule networks in
image classification, their validity in text classification is a relatively recent area of inquiry. The
study achieved an accuracy rate of 74% [15]. The paper introduces the text classification process
and focuses on the CNN model used in this context, which achieved a precision score of 0.85
[16]. Despite numerous proposed approaches, automated text classification continues to be an
active area of research due to the imperfect effectiveness of current classifiers. Notably, few
studies have utilized deep learning techniques for text classification, particularly for non-binary
or multi-category classification problems. Additionally, the LSTM algorithm is a powerful type
of neural network with potential applications in text classification.

Table 1 survy of literature review

Study objective Techniques Accuracy

The purpose of this study is to evaluate
the performance of different deep
learning algorithms in text
Convolutional
classification tasks on medical notes CNN, LSTM,
Neural
Hongxia Lu, 2022 related to various diseases. The dataset RNN, GRU, Bi-
Network 0.
used in the study includes 1,237 unique LSTM
93.
discharge notes, and the task is a
binary-class problem involving
categorization into two classes.
This study employs a feature selection
SVM., NB NB 96.7%,
Thirumoorthy,2022 method that uses the measure of term
classifier SVM 95.1%
frequency distribution.
The objective of this study is to analyze
the impact of n-gram feature selection support Vector
on news article text classification using Machine SVM 93.64
Sudha, D. N. 2021.
semi-supervised learning methods. Naïve Bayes

54
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

The authors suggest using pre-trained

language models and logic rules to create Prompt Tuning
prompts with sub-prompts for multi-class with Rule pre- 91.9 F1
Han, X,2021
trained language scores
text classification. This method is called
models PLMs
prompt tuning with rules.

CNN 91.24%,
This study introduces an integrated LR 89%., NB
Tian, D., 2021 approach that utilizes natural language CNN, LR, NB, 88.32, SVM
processing technology (NLP). SVM, RF, XG 86.67, RF
84.74, XGB
80.48
The authors used the Support Vector
Machines (SVM) model for classifying
Luo, 2021 SVM SVM 90%
English text and documents in this paper.

The study employed the Doc2vec word SVM, GNB, CNN 96.41%,
embedding method for text classification Random Forest GNB 91,48%,
RF 89.72 %,
Dogru,2021 on the Turkish Text Classification (RF), Naive
NB 91,44 %,
dataset. Bayes (NB),
SVM 95,05%
CNN
This paper investigates the potential and Support Vector SVM 95.13,
effectiveness of Machine Learning Machine, RF 0.90, NB
(Tzimourtas,2021)
models in text classification. Random Forest, 0.94
Naive Bayes
hybrid model
gave 96.76%
This study combines two supervised accuracy as
machine learning algorithms, Naïve against the
Bayes and Support Vector Machines hybrid model 61.45% and
(Asogwa,2021) 69.21% of the
(SVM), with text mining techniques to (SVM, NB) Naïve Bayes
create a hybrid model for text and SVM
classification. models
respectively

This research introduces a novel

(Janani, 2021) algorithm, the Optimization Technique MLearn-ATC 0.93
for Feature Selection (OTFS) algorithm,
for text classification.
The paper conducts an empirical
investigation of the utilization of capsule capsule 74%
(Kim,2020).
networks for text classification. networks

55
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

The paper provides an introduction to

(Cai, 2018). text classification and specifically
CNN 0. 85
focuses on the convolutional neural
network.

3 Methodology & material

The main objective of this research is to classify relevant documents based on their content while
minimizing time complexity. The study comprises four primary processes, namely data collection,
data pre-processing, feature extraction, and the application of NLP algorithms. Figure 1 illustrates
the proposed framework for document classification. In the first step, data collection was conducted
from an open-source library such as Kaggle. We selected datasets on biomedical text document
classification gathered from Kaggle. After data collection, the next step is preprocessing, which is
detailed in the methodology section. Once preprocessing is completed, we performed feature
extraction for our experiments. Our chosen features include word frequency, question marks, full
stops, initial words, and final words of the documents. We utilized two classification methods,
Logistic Regression (LR), LSTM, and CNN respectively, for English text mining. Each stage is
described further below.

Data Collection

Data Imbalance

Data Preparation
− Remove Punctuation.
− Transformation
− Stop Word.
− Lemmatization
− Tokenization

Feature Extraction

Data Partition

Text Classification (NLP Technique)

Evaluation Model
Figure 1 Overview of text classification model

56
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

1) Data Collection

For this study, we obtained a dataset on Biomedical text documents, abstracts, and full papers
from the Kaggle website. We used the subset of this dataset that included long research papers
with six or more pages. The dataset consisted of cancer-related documents, which were divided
into three categories: Thyroid Cancer, Colon Cancer, and Lung Cancer. The total number of
publications in the dataset was 7,569, and it contained three class labels. The number of samples
in each category was as follows: Colon Cancer (2,579), Lung Cancer (2,180), and Thyroid Cancer
(2,810).

Table 2 Sample of dataset

2) Data imbalance

When learning from class label-imbalanced data, the prediction model's accuracy may be misleading [17].
Therefore, it is important to assess whether the dataset's class labels are balanced. In our dataset, as shown in
Figure 3, the number of samples for the "Colon" category was 2,579, the number of samples for the "Lung"
category was 2,180, and the number of samples for the "Thyroid" category was 2,810. Based on these
numbers, we can conclude that all categories are balanced.

Data Imbalance

37.1 34.1
Colon Lung Thyroid

28.8

Figure 2 Visualization of Dataset Imbalance

57
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

3) Data preprocessing

Data preprocessing involves preparing and cleaning text to facilitate classification. Text and
document datasets often contain unnecessary words such as stop words, misspellings, and slang.
Noise and irrelevant features can negatively impact system performance, particularly in statistical
and probabilistic learning algorithms [18] [12]. In this section, we employed various techniques and
methods for text cleaning and preprocessing. Basic preprocessing steps were performed on the
datasets before they were vectorized. These steps included the following [19].

3.1 Punctuation Removal

Punctuations such as {"?", "!", ".", ";"} hold no actual importance when it comes to the analysis of
the data. So, the better practice of data analysis involves the removal of punctuation beforehand.

3.2 Text Lowercasing

The text document is transformed into the lowercase so that the uppercase and lowercase
words with same meaning are not treated differently.

3.3 Stop-words Removal.

The process of categorizing texts and documents involves several words that lack
significant meaning for classification algorithms, such as "a", "about", "above", "across",
"after", "afterwards", "again", and so on. To handle these words effectively, they are often
eliminated from the texts and documents using a common approach.

3.4 Text Lemmatization

Lemmatization refers to the process of identifying the root word, also known as the stem,
from various types of words. For instance, words like 'Friendly' and 'Friends' can be
reduced to a common word 'Friend' by using a suffix-stripping algorithm. This technique is
widely used in text classification systems for intelligent information retrieval purposes.

3.5 Text Tokenization

Tokenization is a technique used in pre-processing of text, wherein a continuous stream of
text is divided into smaller units known as tokens, such as words, phrases, symbols, etc.
The primary objective of this step is to analyze the words present in a sentence. For tasks
like text classification and text mining, a parser that processes the tokenization of the
documents is necessary: sentence [20] [21] [2].
After sleeping for four hours, he decided to sleep for another four. In this case, the tokens
are as follows:
{“After” “sleeping” “for” “four” “hours” “he” “decided” “to” “sleep” “for” “another”
“four”}.

58
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

3.6 Feature Extraction

Text classification accuracy is significantly impacted by the process of text feature
extraction. Feature extraction involves treating text as a dot in an N-dimensional space
based on a vector space model [22].
Text Feature Extraction refers to the process of converting text into a set of features
represented by a vector in real number form, which is used as input for classification
purposes. In this study, the process of feature extraction from text utilized the TF-IDF
model [23]. The application of Term Frequency Inverse Document Frequency (TF-IDF)
weighing scheme is one of the approaches to extract features [24]. The TF-IDF weighting
scheme assigns importance to a keyword based on its frequency in a document and
relevance across the corpus. This process involves weighing the keyword in any context,
checking how often it appears in the document and how relevant it is throughout the entire
corpus. [25] [26]. The overall approach of this feature extraction method works as follows.
Given a document collection D, a word w and an individual document d∈ D, we calculate:
𝑤𝑖,𝑗 = 𝑡𝑓𝑖,𝑗 × log(𝑁/𝑑𝑓𝑖 ) (1)
𝑡𝑓𝑖𝑗=number of occurrences of 𝑖 in 𝑗, 𝑑𝑓𝑖=number of documents containing 𝑖, 𝑁=total
number of documents.

4) Data Partition

In order to train the classification model, the dataset is divided into two non-overlapping parts for
training and testing purposes. The hold-out part represents 25% of the dataset and is used for testing,
while the other 75% is used for training. This portion is referred to as "hold-out" since it is held out
for testing, while the remaining data is used to develop the model [27].

Table 3 show data splitting using hold-out validation method.

Category Input Target % Partition

Part 1 6056 (6056, 3) 75% Data Training
Part 2 1514 (1514) 25% Hold-out/ Test set
Colon =514,
Lung=433,
Thyroid =567

59
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

5) Text Classification Algorithms

At this stage, we applied some natural language algorithms in the classification process, namely
Logistic Regression, long short-term memory (LSTM) and CNN. As indicated in figure, the model
was subsequently trained using the training data selected in the previous phase (data splitting)

Documents/
Data Training
Dataset

Rules Text classification models

/knowledge / NLP Algorithms

Figure 3 traning model process

a) Logistic regression

Logistic regression is a supervised classification algorithm that has gained significant importance in
recent times. This algorithm works by taking input and multiplying the input value with weight
value. It is a type of classifier that learns which features from the input are most useful for
distinguishing between different possible classes [28].

b) Convolutional Neural Network

The Convolutional Neural Network (CNN) provides the best feature extraction techniques through
various layers, including embedding, convolutional, pooling, and fully connected layers. The
embedding layer transforms preprocessed text by converting each word into an embedding vector
with a specific dimension. The convolutional layer reduces the dimensionality of features or phrases,
while the pooling layer merges similar features together. The flatten layer then transforms the 2D
features into a vector form suitable for the fully connected layer. Dropout and activation functions
are used in one or more neural layers, called the fully connected or dense layer, to train the model.
The output layer, which has nodes equal to the number of classes in the dataset, uses the SoftMax
activation function to predict models. Another technique used is Long Short-Term Memory (LSTM).
[29].

LSTM has the characteristics of long-distance context dependent learning and can store context
history information; thus, it is used as timing layer of the model in this paper. The module structure

60
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

of LSTM is shown in Fig. 4. [30] [31].

Figure 4 Structure of LSTM

6) Evaluation Model & Results

For multiclass classification, performance measures like Confusion Matrix, Accuracy, Precision,
Recall, and F1-Score are defined based on four features: true positive (TPi), true negative (TNi), false
positive (FPi), and false negative (FNi) of class Ci. If there are m classes in the dataset, i ranges from
1 to m. There are three ways to calculate precision, recall, and F1-score over the entire test data:
macro-averaged, micro-averaged, and weighted-averaged. In this experiment, accuracy, as well as
precision, recall, and F1-score based on weighted-average, are used to evaluate the classifier's
performance. These performance measures are defined as shown in [11] [28][29] [32].

𝑇𝑃+𝑇𝑁 𝑇𝑃
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = (2) 𝑅𝑒𝑐𝑎𝑙𝑙 = (3)
𝑇𝑃+𝑇𝑁+𝐹𝑃+𝐹𝑁 𝑇𝑃+𝐹𝑁

𝑇𝑃 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ∗ 𝑟𝑒𝑐𝑎𝑙𝑙
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = (4) 𝐹1. 𝑆𝑐𝑜𝑟𝑒 = 2 ∗ (5)
𝑇𝑃 + 𝐹𝑃 𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 + 𝑟𝑒𝑐𝑎𝑙𝑙

Figure 5 confusion matrix of LSTM Figure 6 Confusion Matrix of LR

Figure 7 Confusion Matrix of CNN

61
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

Table 3 Accuracy of models

Accuracy 𝑷𝒓𝒆𝒄𝒊𝒔𝒊𝒐𝒏 𝑹𝒆𝒄𝒂𝒍𝒍

LSTM 0.980 0.984 0.982
CNN 0.969 0.965 0.964
LR 0.924 0.925 0.923

Accuracy
100
98
96
94
98
92 96
90 92
88
LSTM CNN LR
LSTM CNN LR

Figure 8 visualization of techniques results

From the results in Table 4 and Figure 8, we have calculated the accuracy precision, and recall value
based on our simulation. The equations are provided in Equations. (2), (3), and (4) respectively. From
Table 4 and Figure it’s obvious that LSTM techniques provide more efficiency as compared to CNN
and Logistic Regression, each technique achieved an accuracy of 0.9940.973, 0.954 respectively.
When comparing the results of the proposed model and previous studies that used the same
techniques, it became clear that the accuracy of the proposed model is higher than the accuracy of the
models of previous studies, as shown in Table 5.

Table 4 comparsion between previous studies and proposed models

Techniques
Studies LSTM CNN LR
Hongxia Lu, 2022 75% 93% -
Tian, D., 2021 - 91.24%, 89%
Dogru,2021 - 96.41%, -
(Cai, 2018). - 85% -
Proposed model 98% 96.9% 92.3%

Conclusion

Discharge medical notes written by physicians that contain important information about the health

62
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

condition of patients. Many deep learning algorithms have been successfully applied to extract
important information from unstructured medical notes data that can entail subsequent actionable
results in the medical domain. This study aims to explore the model performance of various deep
learning algorithms in text classification tasks on medical notes, in this study, we employed 3
artificial intelligence models, CNN (Convolutional Neural Network), LSTM (Long short-term
memory), and Logistic regression. The analysis of these 3 categories of classification problems
showed that the LSTM model performs the best.

REFERENCES

[1] Sarkar, A., & Datta, D. (2017). A Frequency Based Approach to Multi-Class Text
Classification. International Journal of Information Technology and Computer Science (IJITCS), 9(5),
15-22.
[2] Janani, R., & Vijayarani, S. (2021). Automatic text classification using machine learning and
optimization algorithms. Soft Computing, 25(2), 1129-1145.
[3] Lee, J., Song, H. J., Yoon, E., Park, S. B., Park, S. H., Seo, J. W., ... & Choi, J. (2018). Automated
extraction of Biomarker information from pathology reports. BMC medical informatics and decision
making, 18(1), 1-11.
[4] Santos, T., Tariq, A., Gichoya, J. W., Trivedi, H., & Banerjee, I. (2022). Automatic Classification of
Cancer Pathology Reports: A Systematic Review. Journal of Pathology Informatics, 13, 100003.
[5] Zhang, X., Zhang, Y., Zhang, Q., Ren, Y., Qiu, T., Ma, J., & Sun, Q. (2019). Extracting
comprehensive clinical information for breast cancer using deep learning methods. International
Journal of Medical Informatics, 132, 103985.
[6] Lu, H., Ehwerhemuepha, L., & Rakovski, C. (2022). A comparative study on deep learning models for
text classification of unstructured medical notes with various levels of class imbalance. BMC Medical
Research Methodology, 22(1), 181.
[7] Sudha, D. N. (2021). Semi Supervised Multi Text Classifications for Telugu Documents. Turkish
Journal of Computer and Mathematics Education (TURCOMAT), 12(12), 644-648.
[8] Han, X., Zhao, W., Ding, N., Liu, Z., & Sun, M. (2021). Ptr: Prompt tuning with rules for text
classification. arXiv preprint arXiv:2105.11259.
[9] Tian, D., Li, M., Shi, J., Shen, Y., & Han, S. (2021). On-site text classification and knowledge mining
for large-scale projects construction by integrated intelligent approach. Advanced Engineering
Informatics, 49, 101355.
[10] Thirumoorthy, K., & Muneeswaran, K. (2022). Feature selection for text classification using machine
learning approaches. National Academy Science Letters, 45(1), 51-56.
[11] Luo, X. (2021). Efficient english text classification using selected machine learning
techniques. Alexandria Engineering Journal, 60(3), 3401-3409.
[12] Dogru, H. B., Tilki, S., Jamil, A., & Hameed, A. A. (2021, April). Deep learning-based classification
of news texts using doc2vec model. In 2021 1st International Conference on Artificial Intelligence and
Data Analytics (CAIDA) (pp. 91-96). IEEE.
[13] Tzimourtas, A., Bakalakos, S., Tselenti, P., & Voulodimos, A. (2021, November). An exploration on
text classification using machine learning techniques. In 25th Pan-Hellenic Conference on
Informatics (pp. 247-249).
[14] Asogwa, D. C., Anigbogu, S. O., Onyenwe, I. E., & Sani, F. A. (2021). Text classification using
hybrid machine learning algorithms on big data. arXiv preprint arXiv:2103.16624.
[15] Kim, J., Jang, S., Park, E., & Choi, S. (2020). Text classification using
capsules. Neurocomputing, 376, 214-221.

63
 International Journal of Scientific Engineering and Applied Science (IJSEAS) – Volume-9, Issue-6, June 2023
ISSN: 2395-3470
[Link]

[16] Cai, J., Li, J., Li, W., & Wang, J. (2018, December). Deeplearning model used in text classification.
In 2018 15th international computer conference on wavelet active media technology and information
processing (ICCWAMTIP) (pp. 123-126). IEEE.
[17] Garcia, S., Luengo, J., Sáez, J. A., Lopez, V., & Herrera, F. (2012). A survey of discretization
techniques: Taxonomy and empirical analysis in supervised learning. IEEE transactions on Knowledge
and Data Engineering, 25(4), 734-750.
[18] Kowsari, K., Jafari Meimandi, K., Heidarysafa, M., Mendu, S., Barnes, L., & Brown, D. (2019). Text
classification algorithms: A survey. Information, 10(4), 150.
[19] Singh, G., Kumar, B., Gaur, L., & Tyagi, A. (2019, April). Comparison between multinomial and
Bernoulli naïve Bayes for text classification. In 2019 International Conference on Automation,
Computational and Technology Management (ICACTM) (pp. 593-596). IEEE.
[20] Gupta, G.; Malhotra, S. Text Document Tokenization for Word Frequency Count using Rapid Miner
(Taking Resume as an Example). Int. J. Comput. Appl. 2015, 975, 8887.
[21] Verma, T.; Renu, R.; Gaur, D. Tokenization and filtering process in RapidMiner. Int. J. Appl. Inf.
Syst. 2014, 7, 16–18. [CrossRef]
[22] H. Liang, X. Sun, Y. Sun and Y. Gao, "Text feature extraction based on deep learning: a review,"
EURASIP journal on wireless communications and networking, vol. 2017, no 1, 211, 2017.
[23] Indra, S. T., Wikarsa, L., & Turang, R. (2016, October). Using logistic regression method to classify
tweets into the selected topics. In 2016 international conference on advanced computer science and
information systems (icacsis) (pp. 385-390). IEEE.
[24] Dzisevič, R., & Šešok, D. (2019, April). Text classification using different feature extraction
approaches. In 2019 Open Conference of Electrical, Electronic and Information Sciences (eStream)
(pp. 1-4). IEEE.
[25] S. Robertson, "Understanding inverse document frequency: on theoretical arguments for IDF," Journal
of documentation, vol. 60, no. 5, pp. 503-520, 2004
[26] Liu, Q., Wang, J., Zhang, D., Yang, Y., & Wang, N. (2018, December). Text features extraction based
on TF-IDF associating semantic. In 2018 IEEE 4th International Conference on Computer and
Communications (ICCC) (pp. 2338-2343). IEEE.
[27] Yadav, Sanjay, and Sanyam Shukla. "Analysis of k-fold cross-validation over holdout validation on colossal
datasets for quality classification." 2016 IEEE 6th International conference on advanced computing (IACC).
IEEE, 2016.
[28] Shah, K., Patel, H., Sanghvi, D., & Shah, M. (2020). A comparative analysis of logistic regression, random forest
and KNN models for the text classification. Augmented Human Research, 5(1), 1-16.
[29] Behera, B., Kumaravelan, G., & Kumar, P. (2019, December). Performance evaluation of deep learning
algorithms in biomedical document classification. In 2019 11th international conference on advanced computing
(ICoAC) (pp. 220-224). IEEE
[30] Zhou, C., Sun, C., Liu, Z., & Lau, F. (2015). A C-LSTM neural network for text classification. arXiv preprint
arXiv:1511.08630.
[31] Bai, X. (2018, September). Text classification based on LSTM and attention. In 2018 Thirteenth International
Conference on Digital Information Management (ICDIM) (pp. 29-32). IEEE
[32] Ikonomakis, M., Kotsiantis, S., & Tampakas, V. (2005). Text classification using machine learning techniques.
WSEAS transactions on computers, 4(8), 966-974.

64