An in-depth review on News Classification through NLP

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 11 | Nov 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 717
An in-depth review on News Classification through NLP
Labade Saurabh Ravindra1, Thorve Rushikesh Vallabh2, Wavhal Ruturaj Sagar3,
Prof. M. G. Sinalkar4
1,2,3Department of Computer Engineering, Jaihind College of Engineering, Pune, Maharashtra, India
4Asst. Professor, Department of Computer Engineering, Jaihind College of Engineering, Pune, Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - An ever-increasing numberofpeopleintheworld
depend on online news outlets as their major source of
everyday information and current events. As more people
discover how useful digital data can be, both the amount of
information available and the frequency with which it is
accessed are anticipated to grow considerably. With the
massive amounts of data being produced by a plethora of
publishers, it may be challenging for average consumers to
obtain the data they need. Unfortunately, meanwhile, current
search engines return so many items that only asmall fraction
of them are relevant to user requests. As a result, it might be
helpful to layer search engines with a classifier, any algorithm
created to sort large amounts of data into predetermined
categories. There have been considerable inconsistencies that
are observed with the realization of an effective news
classification approach in the current methodologies which
are listed in this survey paper. Thus, to improve this condition
there is a need for an effective and useful news classification
approach that utilizes Natural Language processing and
feature extraction along with Decision Making and fuzzy list.
This approach will well defined in the next research on this
paradigm.
Key Words: Natural Language Processing, Feature
Extraction, Decision Making, Fuzzy List.
1. INTRODUCTION
As computing power and connectivity have
improved over the last several years, so too has the volume
of online data. Getting one's news these days mostly comes
from the big news portals. A growing amount of news
content, unfortunately, poses significantdifficultiesfornews
portals. The requirements of modern society cannot be
satisfied by using the same text categorization techniques
that have been used for decades. As a result, text
categorization modeling development has become more
important in the area of knowledge discovery in recent
times. The speed and accuracy with which the news text
categorization technique processes all text input and
predicts categorization labels is impressive. As a result,
automated classification may provide a cost-effective
solution for completing the text categorization job for the
media outlet. The study of automated text categorization is
becoming more significant in the age of big data.
In today's information age, there are many online resources
like Yandex, Bing, and others from which anybody may
access a wide range of data. In general, the content of such
portals is organized into distinct subcategories for the
convenience of viewers. Anyone may quickly and easily get
the specific pieces of news and data that most interest them.
Included here might be "economic," "educational," "sports,"
and so on. There are numerous news stories in various
categories that are completely unrelatedtothetopicathand.
In recent years, several studies have been presented for the
purpose of news categorization. Researchers in this area
have used a wide range of taxonomical approaches to
analyses their native tongue.
As more people discover how useful digital
information can be, both the volume of information and the
frequency with which it is accessed are expected to soar.
With the massive amounts of data being produced by a
plethora of publishers, it may be challenging for average
consumers to get the information they need. Unfortunately,
nevertheless, current search engines return so many items
that only a small fraction of these are relevant to user
queries. As a result, it might be helpful to layer search
engines with a classification model, an algorithm created to
sort large amounts of data into predetermined categories.
Multiple approaches exist that can accurately categorize an
English text. The majority of these algorithms are broken
down into four sub-categories: text pre-processing, feature
extraction, categorization, and effectiveness.
One of the most critical and difficult problems in
machine learning techniques is text categorization.
Allocating a label from a collection of possible ones to a
document corpus autonomously is the task known as text
categorization. The significance of this document is
determined by the labels that will beapplied.Forthisreason,
even for an ordinary human, picking the right collection of
labels could be a matter of some ambiguity. The papers may
be categorized into one or more categories according to the
information we know about them. The documents in this
data corpus must only be filed under one of the available
categories.
X. Liang et al. present a basic method for identifying
CM articles using TF-IDF characteristics, with only middling
success [1]. Finally, a graph-based approach is proposed to
further improve the identification of content marketing

pieces. To begin, two distinct graph types are developed,
each made up of a Sentence Graph and a Word Graph. The
authors then offer an innovative technique for determining
Sentence Graph edge weights that take into account both
semantic similarity and temporal correlations between
sentences. Features linked to both graphs and communities
may be derived from these two types of graphs. Next, the
authors use a dataset that has been manuallylabeledtotrain
a supervised classifier. The evaluation experiments also
reveal that the graph-based features extracted from the
graphs can vastly outperform the control group's methods.
The WVDD text similarity measure strategy was
introduced by S. Zhou et al. To further increase the rate at
which text is clustered, the authors implemented the K-
means algorithm using the Spark architecture, which allows
for parallel computation. The benefits and viability of the
suggested technique are confirmed by F-measure
experimental verification, where it is contrasted with the
Doc2vec model and the Bag-of-words model [2]. Evidence
from experiments shows that WVDD takes into account the
structure, word order, and weight settings of Chinese
sentences. The suggested technique is more applicable and
takes into account semantic relations when the text dataset
has a standardized, Chinese sentence component structure
and the sentence length is short. Furtherstudies,suchastext
sentiment analysis, information retrieval, the semantic
cognition of artificial intelligence, and so on, might be
conducted once this technique yielded more accurate
findings for text similarity. When applied to text similarity
assessment, a fundamental tool in big data research, the
suggested technique can improve the efficiency of
information mining.
In this article that analyzes literature, chapter 2 is
broken up into an assessment of previous research that is
presented in the manner of a reviewed literature, and the
third chapter concludes with somesuggestionsforhowmore
research should be conducted.
2. RELATED WORKS
A hybrid bidirectional recurrent convolutional
neural network attention-based model (BRCAN) has been
proposed for fine-grained text categorization by J. Zheng et
al. This model integrates the Bi-LSTM and CNN efficiently
with the aid of the word2vec model and attention
mechanism [3]. The proposed model, as is well known, has
many benefits, including the following: it captures the
contextual information and the semantics of long text by Bi-
LSTM to alleviate the problem of information imbalance and
save the time-step information; it selects higher-level local
features useful for classification from the intermediate
sentence representation generated by Bi-LSTM according to
the context generated by CNN, and fewer parameters are
used to obtain the interactionbetweenhiddenlayer.Because
of this, the proposed model takes the best features from
three different models to create a unified representation ofa
text. The authors compare the proposed model to state-of-
the-art classification models that depend on conventional
machine learning and deep learning techniquesandvalidate
it on multi-topic classification and fine-grained sentiment
analysis tasks.
A novel WSD model, proposed by Y. Heo et al.,
accounts for the increased uncertainty around the senses of
some words while also displaying improved prediction
accuracy for uncommon senses. Dependingonthecontext of
the sample phrase, the model will choose one of several
alternative interpretations of the target word. The structure
of the Oxford Dictionary inspires this;there,eachmeaning of
a word is recorded and then organized based on its assigned
part of speech [4]. Words that have several meanings might
benefit from this technique since the scope of possiblesense
choices is narrowed down. As a bonus, neural language
models may now generate unique pieces of context for the
target word depending on the segment of speech it is in. The
authors also propose a hybrid sense prediction system in
which less common word senses are categorized
independently from more common ones. This improves
prediction accuracy, even if there is significant ambiguity in
the sensory environment when working with a small
number of training phrases.
W. Zhao et al. present an approach to text
categorization that uses only a subset of the availabletext. In
today's online environment, it's usual to encounter financial
writings with inconsistent information quality and
incomplete textual content. Partial-text-based text
categorization may so mimic the incomplete-information
environment and is thus more applicable to real-world
scenarios. Text classification of subsets of text is more
demanding and difficult than text classification of the
complete text [5]. The authors propose a neural network,
AD-CharCGNN that takes advantage of both charCNN and
GRU. In other words, the AD-CharCGNN can glean data from
both the temporal and geographical dimensions. Character
level data drives the AD-CharCGNN. Whether in Chinese,
English, numerals, or other unique characters, messagesare
surprisingly common. All of the aforementioned characters
are supported by the network,anda character-level network
may read the whole data set without first having to filter out
unnecessary words.
Optimized machine learning and deep learning
techniques to identify false news were proposed by D. S.
Abdelminaam et al. using COVID-19 and other data sets.
Tokenization and stemming were just two parts of a
comprehensive sentence analysis that took place in the
preparatory phase [6]. There are three datasetsintotal,with
one utilized for training and the othertwofortesting.TF-IDF
and Ngrams are used in the machine learning method to
feature analysis, whereas word embedding is used in the
deep learning technique. Grid search and Keras tuning are

used to achieve the best possible results from each method.
Accuracy, precision, recall, and the F1 measure are all
utilized to evaluate the effectiveness of both methods.
D. Alsaleh et al. suggested a CNN-GA mixed
classification model for Arabic text. Two big datasets were
used to verify the suggested model. The outcomes of using
GA-CNN were fantastic. In addition, the model fared well in
head-to-head tests against both the baseline and a
preexisting technique [7]. Therefore, the improvement in
classification accuracy and RMSE for Arabic textachieved by
combining CNN with GA was confirmed. Since GA is run
throughout the training and validation phase to get the
optimal weights, GA-CNN requires more time to calculate
than the baseline method.
Two methods, traditional ML and DL, were used by
H. Saleh et al. to create a system for detectingbogusnews. As
the top-performing DL model, the suggested OPCNN-FAKE
model has been implemented. Embedding, dropout,
convolutional, pooling, flattening,andoutputlayersmake up
the six-layer suggested OPCNN-FAKE model [8]. Also, it has
been fine-tuned with the help of the hyperopt optimization
method, wherein the authors triedouta varietyofparameter
settings for each layer and settled on the optimal
combination. Similar to how n-grams with TF-ID have been
employed in ML and DL, word embedding feature extraction
approaches have also been used.
S. A. Sulaimani et al. [9] propose a new method for
multi-class text classification that makes use of Contextual
Analysis. Multiple experimentswereconstructedtoevaluate
the effectiveness of the proposed method under two
conditions: Unbalanced Classes and a Large Number of
Classes. Naive Bayes, Support Vector Machines, K-Nearest
Neighbors, and Convolutional Neural Networksareusedona
Twitter event corpus to compare their performancetoother
popular classification methods. With an average f1>97.09%
and f1>95.27% in the imbalanced classes and the high
number of classes experiments, respectively, the results
show that the proposed method performs well in classifying
brief texts (tweets) into various groups (events). This level
of efficiency is generally accepted as adequate for most
projects. The interpretability analysis also shows that this
technique is straightforwardtounderstandincomparison to
the others employed in this research.
A large-scale unsupervised Bangla languagedataset
(BanglaLM) is proposed for linguistic research by M.
Kowsher et al. This work covers investigating the feasibility
of fine-tuning a transformer model for a low-resource
language like Bangla and traininga languagemodel using the
biggest dataset yet produced for Bangla [10]. With this
research, the mBERT's mixed weights problem for 104
languages, including Bangla (trained on restricted and more
structured data only), is fixed. The authors take a look at
Bangla-BERT and demonstrate its efficiencyacrossfourNLP
downstream tasks: sentiment analysis, named entity
recognition, binary, and multilevel text classifications.
Finally, they demonstrated that the proposed model
significantlyoutperformed mBERTandothernon-contextual
models in various downstream tasks, like Bangla fasttext,
and word2vec.
D. Jung et al. suggested a news-specific model for
detecting document inconsistencies using a graph-based
approach. By analyzing textual information, GraDID can
determine if a given body context is consistent or not. The
authors used two tasks in English and Korean to assess the
GraDID. UsingNELA17,they improveuponpreviousstate-of-
the-art results in the area of inconsistent document
detection [11]. They also demonstrate how supernode may
be utilized for direct categorization due to its ability to
effectively collect the whole information. The authors think
that method has wide-ranging potential, from assessing the
quality of body text to identifying fake news.
A multi-state support vector machine (MSVM)
method was discussed by Ravish et al. By employing a
distinct model for feature selection and feature extraction,
they have been able to fine-tune the model fora widevariety
of datasets. The authors employed an improved version of
PCA with two phases, dubbed TP-PCA, to extract features.
For the Multi-Class Support Vector Machine, they also
suggested a feature extraction approach based on Firefly
[12]. The suggested technique has been evaluated on 10
various datasets, such as FakeNewsNet, LIAR, ISOT,
PolitiFact, etc. Due to a large number of features present in
the datasets, feature selection approaches proved more
effective than those that employed all of the features during
deep learning model training. The feature extraction
algorithms performed poorly only on datasets with a low
feature count.
Fig. System Overview Diagram
3. CONCLUSION
Both the amount of data stored and the number of
times it is retrieved are projected to increasedramaticallyas
more individuals learn to use the benefits of online

resources. Users may find it difficulttoevengetthedata they
require in the ocean of information getting created by a
proliferation of sources. However, modern search engine
crawlers provide such a massive list of results that even a
tiny percentage of them are really appropriate to users'
searches. Therefore, it may well be useful to overlay search
results with a categorization framework, a technique
developed to classify enormous volumes of informationinto
designated groups. There are a number of methods that can
successfully classify an English text. This research surveys
the current state of the art in news classification techniques
and finds that there are many discrepancies that have been
discovered with the achievement of a successful news
classification methodology. As a result, there is a need for a
powerful and practical news categorization strategy that
makes use of Natural Language processing, feature
extraction, Decision Making, and fuzzy liststoamelioratethe
current situation. In subsequentstudiesusingthisparadigm,
this method will be more clearly described.
REFERENCES
[1] X. Liang, C. Wang, and G. Zhao, "Enhancing Content
Marketing Article Detection With Graph Analysis," in
IEEE Access, vol. 7, pp. 94869-94881, 2019, DOI:
10.1109/ACCESS.2019.2928094.
[2] S. Zhou, X. Xu, Y. Liu, R. Chang, and Y. Xiao, "Text
Similarity Measurement of Semantic CognitionBasedon
Word Vector Distance Decentralization With Clustering
Analysis," in IEEE Access, vol. 7, pp. 107247-107258,
2019, DOI: 10.1109/ACCESS.2019.2932334.
[3] J. Zheng and L. Zheng, "A Hybrid Bidirectional Recurrent
Convolutional Neural Network Attention-Based Model
for Text Classification," in IEEE Access, vol. 7, pp.
106673-106685, 2019, DOI:
10.1109/ACCESS.2019.2932619.
[4] Y. Heo, S. Kang and J. Seo, "Hybrid Sense Classification
Method for Large-Scale Word Sense Disambiguation,"in
IEEE Access, vol. 8, pp. 27247-27256, 2020, DOI:
10.1109/ACCESS.2020.2970436.
[5] W. Zhao, G. Zhang, G. Yuan, J. Liu, H. Shan, and S. Zhang,
"The Study on the Text Classification for Financial News
Based on Partial Information," in IEEE Access, vol. 8, pp.
100426-100437, 2020, DOI:
10.1109/ACCESS.2020.2997969.
[6] D. S. Abdelminaam, F. H. Ismail, M. Taha, A. Taha, E. H.
Houssein and A. Nabil, "CoAID-DEEP: An Optimized
Intelligent Framework for Automated DetectingCOVID-
19 Misleading Information on Twitter," in IEEE Access,
vol. 9, pp. 27840-27867, 2021, DOI:
10.1109/ACCESS.2021.3058066.
[7] D. Alsaleh and S. Larabi-Marie-Sainte, "Arabic Text
Classification Using Convolutional Neural Network and
Genetic Algorithms," in IEEE Access, vol. 9, pp. 91670-
91685, 2021, DOI: 10.1109/ACCESS.2021.3091376.
[8] H. Saleh, A. Alharbi and S. H. Alsamhi, "OPCNN-FAKE:
Optimized Convolutional Neural Network forFakeNews
Detection," in IEEE Access, vol. 9, pp. 129471-129489,
2021, DOI: 10.1109/ACCESS.2021.3112806.
[9] S. A. Sulaimani and A. Starkey, "Short Text Classification
Using Contextual Analysis," in IEEE Access, vol. 9, pp.
149619-149629, 2021, DOI:
10.1109/ACCESS.2021.3125768.
[10] M. Kowsher, A. A. Sami, N. J. Prottasha, M. S. Arefin, P. K.
Dhar and T. Koshiba, "Bangla-BERT:Transformer-Based
Efficient Model for Transfer Learning and Language
Understanding," in IEEE Access, vol. 10, pp. 91855-
91870, 2022, DOI: 10.1109/ACCESS.2022.3197662.
[11] D. Jung, M. Kim and Y. -S. Cho, "Detecting Documents
With Inconsistent Context," in IEEE Access, vol. 10, pp.
98970-98980, 2022, DOI:
10.1109/ACCESS.2022.3204151.
[12] Ravish, R. Katarya, D. Dahiya, and S. Checker, "Fake
News Detection System Using Featured-Based
Optimized MSVM Classification," in IEEE Access, vol.10,
pp. 113184-113199, 2022, DOI:
10.1109/ACCESS.2022.3216892.

An in-depth review on News Classification through NLP

Recommended

More Related Content

Similar to An in-depth review on News Classification through NLP (20)

More from IRJET Journal (20)

Recently uploaded (20)

An in-depth review on News Classification through NLP