Deep Learning Algorithms for Detecting Fake News in Online Text

Abstract

Spread of fake news is a social phenomenon
that is pervasive at the social level between individuals,
and also through social media such as facebook and
twitter. Fake news that we are interseted about is one of
many kinds of deception on social media , but it’s more
important one as It is created with dishonest intention to
mislead peoples. We are concerned about this issue
because we have noticed that this phenomenon has
recently caused through the means of social
communication to change the course of society and
peoples and also their views, for example, during
revolutions in some Arab countries have emerged some
false news that led to the absence of truth and stir up
public opinion and also fake of news is one of the factors
Trump success in the presidential election. So we decided
to face and reduce this phenomenon, which is still the
main factor to choose most of our decisions.Techniques of
fake news detection varied, ingenious, and often exciting.
In this paper our objective is to build a classifier that can
predict whether a piece of news is fake or not based only
its content, thereby approaching the problem from a
purely deep learning perspective by RNN technique

models (vanilla ,GRU) and LSTMs . We will show the
difference and analysis of results by applying them to
dataset that we used called LAIR .we found that the
results are close, but the GRU is the best of our results
that reached (0.217) followed by LSTM (0.2166) and
finally comes vanilla (0.215) . Due to these results we will
seek to increase accuracy by hybrid between GRU and
CNN techniques on the same dataset.

Full text

Deep Learning Algorithms for Detecting Fake
News in Online Text
Sherry Girgis
Faculty of Computer Science
Modern Academy for Computer Science
and Management Technology
Cairo, Egypt
sherrygirgis91@gmail.com
Eslam Amer
Faculty of Computer Science
Misr International University
Cairo, Egypt
eslam.amer@miuegypt.edu.eg
Mahmoud Gadallah
Faculty of Computer Science
Modern Academy for Computer Science
and Management Technology
Cairo, Egypt
mgadallah1956@gmail.com
Abstract— Spreading of fake news is a social phenomenon
that is pervasive at the social level between individuals, and also
through social media such as Facebook and Twitter. Fake news
that we are interested in is one of many kinds of deception in
social media, but it’s more important one as it is created with
dishonest intention to mislead people. We are concerned about
this issue because we have noticed that this phenomenon has
recently caused through the means of social communication to
change the course of society and peoples and also their views, for
example, during revolutions in some Arab countries have
emerged some false news that led to the absence of truth and stirs
up public opinion and also fake of news is one of the factors
Trump successes in the presidential election. So we decided to
face and reduce this phenomenon, which is still the main factor to
choose most of our decisions. Techniques of fake news detection
varied, ingenious, and often exciting. In this paper our objective
is to build a classifier that can predict whether a piece of news is
fake or not based only its content, thereby approaching the
problem from a purely deep learning perspective by RNN
technique models (vanilla, GRU) and LSTMs. We will show the
difference and analysis of results by applying them to the dataset
that we used called LAIR. We found that the results are close,
but the GRU is the best of our results that reached (0.217)
followed by LSTM (0.2166) and finally comes vanilla (0.215). Due
to these results, we will seek to increase accuracy by applying a
hybrid model between the GRU and CNN techniques on the same
data set.
Keywords— Deception detection; Deep Learning; Artificial
Intelligence; RNN (Recurrent Neural Network); LSTM (long
short-term memories) ; Vanilla ; GRU (Gated Recurrent Unit) ;
CNN(Convolutional Neural Networks ).
Keywords— Deception detection; Deep Learning; Artificial
Intelligence; RNN (Recurrent Neural Network); LSTM (long
short-term memories); Vanilla; GRU (Gated Recurrent Unit);
CNN(Convolutional Neural Networks ).
I. INTRODUCTION
Social media for news consumption, such as Facebook and
Twitter is a double-edged sword [1]. On the one hand, low
cost and easy access to information and dissemination quickly
to push people to search for news and know what is happening
at the beginning of events with details and updates at the
moment unlike newspapers or magazines in the old days, on
the other hand, it enables the widespread of 'fake news'
because of its accessibility and lack of cost and control of the
Internet. Recent reports suggest that the outcome of the U.S.
Presidential Elections is due to the rise of online fake news.
Reports indicate that the human ability to detect deception
without special assistance is only 54%. So we need to use
machine learning for classifying texts automatically [2].
Fake news detection is considered one of the most
dangerous types of deception because it recently caused
deceiving many people, Fake news defined as the prediction of
the chances of a particular news article (news report, editorial,
expose, etc.) being intentionally deceptive (Rubin, Conroy &
Chen, 2015) [3].
We concerned about the fake news because of the problem
of fake news detection is more challenging than detecting
deceptive reviews. A recent report by the Jump-shot Tech
Blog1found that Facebook referrals accounted for 50% of the
total traffic to fake news sites and 20% of the total traffic to
reputable websites [4]. Since the majority of U.S. Adults –
62%– gets news on social media (Jeffrey and Elisa, 2016) [5]
so that the ability to identify fake content in online sources is,
therefore, an urgent need.
Our system establishes for detecting fake news by using
deep learning technique that shows an improvement than
linguistic cues recently, we use RNN models (vanilla, and
GRU) and LSTM technique with the LAIR dataset (discuss it
in details in next sections).
The organization of the paper is as follows: The next
section we will talk about the fake news problem in general
and changelings that faced us and the researchers when trying
to reach the best result, then will presenting the previous
findings of the researchers in this field to classify news
passages as fake or not, next we will mention our experiments
in details that we reached with the dataset that we use, and it's
preparation, next we will mention our view of the steps of the
system and proposed our model, Then will clarify our results
93
978-1-5386-5110-0/18/$31.00 ©2018 IEEE

and analysis of each result, Finally will presenting conclusions
of all sections.
II. PROBLEM DEFINITION
Fake news becomes a major issue for the public and
government. Fake news can take advantage of multimedia
content to mislead readers and get published, which can lead
to negative effects or even manipulation of public events. One
of the unique challenges of detecting fake news on social
media is how to identify fake news about recent events.
The task of detecting fake news has tested a variety of
labels, from misinformation to rumors; to spam. There has
been a large body of work surrounding text analysis of fake
news and similar topics such as rumors or spam. We have tried
to mention some papers that interested in this subject also and
each paper will recall it‟s experience and results and is based
on its own concept of these words deceptive [6].
There are two directions to detect text deception. The first of
which is based on separate handheld features, which can
capture linguistic and psychological causes, however, these
features failed to classify text well, which limits performance.
The second approach based on a neural network model learns
document-level representation to discover deception text.
Neural network models have been used to learn semantic
representations for NLP tasks and it reaches the highest
competitive result According to (Le and Mikolov, 2014; Tang
et al., 2015) [7] and also in NLP, fake data have been collected
by crawling the web or crowdsourcing: fake product reviews
(Mukherjee, Venkataraman, Liu, & Glance, 2013) [8].
Most researchers used many deep learning popular
algorithms such as CNN, Bidirectional-LSTM, and RNN
that we will mention in detail later, but datasets was a reason
that why the fake news detection was not successful in the past
because it was small and included unrealistic news, In 2017
William created a new benchmark dataset called
LAIR which Collected 12.8K short data is labeled in different
Contacts from POLITIFACT.COM, which Provides detailed
analysis report and links to source documents for each case.
William works on LAIR dataset with many techniques such as
logistic regression, support vector machines, and
(Bidirectional-LSTM and CNN) models for deep learning and
CNN results is the best [9].
Our work in this paper will implement RNN
models (Vanilla RNN, GRU) and LSTMs on LAIR dataset to
determine if the news is truthful or deceptive and will show
the results compared to William's results with analysis.
The following part, we represent some of the previous
practical neural network models and their datasets to detect
deception on a text.
III. RELATED WORK
Deception detection (in the framework of computational
linguistics) is a text classification problem where our system
should classify an unseen document as either truthful or
deceptive. Such a system is first trained on known instances of
deception. Used features are token unigrams and linguistic
cues derived from the classes of words from the Linguistic
Inquiry and Word Count (LIWC) [10]. This was
a psychological experiment with the analysis of the
participants‟ writings, focusing on the connection between
deception and fantasy proneness. The task of detection that
was performing is thus opinion spam detection or fake news
detection, which is a more variant of deception detection.
In the past in deception detection mostly relied on manual
feature selection based on, for example, psycholinguistic
theories of deception and/or computational linguistics,
followed by supervised machine learning to build a classifier
[11], but recent NLP researches are now increasingly focusing
on the use of new deep learning methods shown in Fig.1 [12].
Fig. 1. Percentage of deep learning papers in ACL, EMNLP, EACL, NAACL
over the last 6 years (long papers)
Here we mention some of the previous works used many
techniques for fake news detection.
(Niall J. Conroy, 2015) [4] Using linguistic cues
approaches and network analysis approaches to design a basic
fake news detector which provides high accuracy in terms of
classification tasks. They propose a hybrid system whose
features like multi-layer linguistic processing, the addition of
network behavior, they propose a method to detect online
deceptive test by using a logistic regression classifier which is
based on POS tags extracted from a corpus deceptive and
truthful texts and achieves an accuracy of 72% which could be
further improved by performing cross-corpus analysis of
classification models and reducing the size of the input feature
vector ,it‟s one of the best results that these features can
reach so as to improve the results the next researchers have to
use neural networks.
(Samir Bajaj, 2017)[13] used many techniques for neural
networks and machine learning to determine which algorithm
reach the best result among of them, he apply these algorithms
to his dataset which collect it two different sources; from an
open Kaggle dataset sold 13,000 Fake news articles and
50,000 authentic news articles (negative examples for the
classifier) were extracted from the Signal Media News dataset.
Split all of these data (63000 articles) into 60% training, 20%
dev/validation, and 20% test sets. He used many techniques
from machine learning and neural network such as Logistic
Regression, Feedforward Network, RNN (Vanilla, GRU),
LSTMs, Bi- LSTMs, CNN with Max-Pooling and CNN with
Max-Pooling and Attention, The results proved that GRU gets
the better F1 score and best results overall.
94

Other research by (Natali Ruchansky et. Al, 2017) [14], for
getting better result more than previous one; she proposed a
model called CSI, it is built up from deep neural networks
which can extract information from different domains and
capture temporal dependencies in user engagement with
articles, and also select important features. CSI (which is
composed of three modules: Capture, Score, and Integrate)
evades the cost of manual features election by incorporating
neural networks. The features they use to capture the temporal
behavior and textual content in a general way that does not
depend on the data context nor requires distributional
assumptions. They created this technique to solve three main
problems in fake text, first of them evaluate the matching
score for the headlines and body of an article, second the
emotion that reach to readers from an article and how make
them feel, the last characteristic is knowing the source of the
article by checking the structure of the URL, to the credibility
of the media source. They used two datasets from Twitter
and Weibo (real-world social media datasets), CSI gives the
best performance overall comparison models and versions. We
see that integrating user features boosts the overall numbers up
to 4.3% from GRU-2. Put together, these results demonstrate
that CSI successfully captures and leverages all three
characteristics of text, response, and source, for accurately
classifying fake news (Shown in table 1).
Other Works by (William Yang et.al, 2017) [12] Due to
lack of fake news datasets and lack of their efficiency, he
decided to present a new benchmark dataset called LIAR (will
mention it in details in next sections): it‟s a new, publicly
available dataset for fake news detection. Use it with many
techniques such as logistic regression, support vector
machines, and (Bidirectional-LSTM and CNN) models for
deep learning. The results proved that CNN
(Convolutional Neural Networks) models are the best.
As we showed that William‟s works did not recover using
RNN technique, although GRU offered the best results,
therefore we use this technique in LAIR dataset and will
compare our works with his results and give a brief analysis.
TABLE I. COMPARISON OF DETECTION ACCURACY ON TWO DATASETS
Twitter
WEIBO
Accuracy
F-score
Accuracy
F-score
DT-RANK
0.624
0.636
0.732
0.726
DTC
0.711
0.702
0.831
0.831
SVM-TS
0.767
0.773
0.857
0.861
LSTM-1
0.814
0.808
0.896
0.913
GRU-2
0.835
0.830
0.910
0.914
CI
0.847
0.846
0.928
0.927
CI-T
0.854
0.848
0.939
0.940
CSI
0.892
8.894
0.953
0.954
IV. EXPERMINENTS
A. Dataset
There are very useful datasets to study fake detection but
the positive training data are collected from a tested out (in a
way that was close to the real thing) (surrounding
conditions). More importantly, these datasets are not good for
fake statements detection; since the fake
news on TVs and social media are much shorter than customer
reviews.
According to (William Yang et.al, 2017) presents a new
benchmark dataset called LIAR: it‟s a new, publicly available
dataset for fake news detection. Collected a decade-long,
12.8K manually labeled short statements in various contexts
from POLITIFACT.COM, Which provides a detailed
analytical report and a link to its source level for each case.
This dataset can be used for fact-checking research as well.
They investigate the automatic detection of fake news based
on surface-level linguistic patterns [12].
The LIAR dataset includes 12,836 short statements labeled
for truthfulness, subject, context/venue, speaker, state, party,
and prior history. With this size and time span of ten years,
LIAR cases are collected in more natural context, such as
political debate, TV ads, Facebook posts, tweets, interview,
news release, etc. In each case, the labeler provides a lengthy
analysis report to the ground each judgment [12].
They have evaluated several popular learning based
methods on this dataset. The baselines include logistic
regression, support vector machines, LSTM and the CNN
model. We will present some examples of LIAR dataset in
Fig.2 [12].
Statement: “Newly Elected
Republican Senators Sign
Pledge to Eliminate Food
Stamp Program in 2015.”
Speaker: Facebook posts
Context: social media
posting
Label: Pants on Fire
Justification: More than
115,000 social media users
passed along a story
headlined, "Newly Elected
Republican Senators Sign
Pledge to Eliminate Food
Stamp Program in 2015."
But they failed to do due
diligence and were
snookered since the story
came from a publication that
bills itself (quietly) as a
"satirical, parody website."
We rate the claim Pants on
Fire.
Statement: "Under the
health care law, everybody
will have lower rates, better
quality care, and better
access." 
Speaker: Nancy Pelosi
Context: on ‟Meet the
Press‟
Label: False
Justification: Even the
study that Pelosi‟s staff cited
as the source of that
statement suggested that
some people would pay
more for health insurance.
Analysis at the state level
found the same thing. The
general understanding of the
word “everybody” is every
person. The predictions
don‟t back that up. We rule
this statement False.
Fig.2. depicts some example from LIAR dataset.
95

B. Data Preparation
After we have LIAR data set as we mention in the previous
section, we must preprocess it to be suitable for our system
that can work with. Preprocessing means that data set is
clearer to our algorithm by removing dummy characters,
string, and Impurities. Preprocessing Works on three steps :
 Splitting: Separate each sentence from the next sentences
to deal with them individually.
 Stop words removing: remove UN-important words from
each sentence.
 Stemming: Returns each word to its origin.
C. Deception System
Due to the deception phenomenon that spread nowadays
through the traditional media and the social media platform
especially the Facebook and Twitter, which controls the users'
selections in relation to the economic part such as (buying
products, business, etc…) also in relation to the political life,
so we decided to create s system which solves this problem
not only fake reviews but we concerned about fake news by
using preprocessing LAIR data which has been prepared based
on what we have cleared in details in the previous part we
apply it to Word embedding (or word vector) that gives each
word a vector and each vector represents a latent feature of a
word , then the result of the word vector apply to RNN models
(Vanilla, GRU) and LSTM , then we will get the results that
determine that a piece of news is deceptive or not.
We will explain in details each step of our system in the
next section.
V. PROPOSED MODEL
Our work clarified in the following steps as follows:
First step: Preparing LIAR dataset in four levels:
 The first level is splitting each sentence to deal with
separately.
 The second level is removing stop words and that
includes identifying the useless words in each statement
like (the, a, an, etc).
 The third level is stemming which every word return to
its infinitive.
Second step: Output of stemming will be the input to word
embedding which played an important role in deep learning
based on deception analysis that includes representing each
single word in each sentence by dimensional vector and get
the relation between two words not only syntactic but also the
same (as „see‟ and „watch‟ are very different in syntactic, but
their meaning is somewhat related) [15]. Another benefit is
that the algorithm detects the words that appear mostly
together (like „wear‟ and „clothes‟) and it shows their
relationship and then this is able to predict the next word [16].
Third step: Results of word embedding level will be the
input to the RNN models (vanilla, GRU ) and LSTMs
technique.
Fourth step: the output of step four will Getting final
result determining if the piece of news is truthful or deceptive.
As is common in data mining problems, once the models are
built, the process might be repeated with new data and new
features.
VI. RESULTS AND ANALYSIS
We constructed three different experiments.
 Vanilla: Is the first model of RNN that used from 1980,
and it‟s just Single Layer Network (with feedback).
 GRU (Gated Recurrent Unit). researchers used to use it
(from 2014) because of avoiding vanilla issues which
filter the information flow to enable the modeling of
long-term dependencies
 LSTM (long short-term memories) Behave like RNNs,
but LSTMs have a different function of computing the
hidden state by introducing input, forget, and output
gate mechanism and an additional memory cell state
(can store information for a longer time).
In the first experiment we used Vanilla RNN, Then we
used GRU model in the second experiment, finally, in the last
experiment we used the LSTM technique. The following table
illustrates the accuracy of each model.
TABLE II. COMPARISON OF OUR RESULTS AND WILLIAM'S ACCURACY
Model
Test Accuracy
SVMs
0.255
Logistic Regress0ion
0.247
Bi-LSTMs
0.233
CNN
0.270
Vanilla
0.215
GRU
0.217
LSTM
0.2166
We compared our results with (William Yang et.al, 2017)
in the previous table II.
We found the worst result of our experiments gets from
vanilla because of its failure to solve complex tasks that have a
practical application, It also changed the format of the original
information, which meant that it was unable to hold the
important memory content for more than a few time steps,
And also Gradient vanishing Is one of its disadvantages.
LSTM also showed inefficiency compared to GRU and
CNN because its two main drawbacks, first it is more
expensive to calculate the network output and apply back
propagation. we simply have more maths to do because of the
complex activation. However this is not as important as the
second point, second, the explicit memory adds several more
weights to each node, all of which must be trained. This
increases the dimensionality of the problem and potentially
makes it harder to find an optimal solution.
The best result of our experiments is GRU, we have seen a
slight improvement in its results than vanilla and LSTM
because of solving gradient vanishing problem which is a
problem in vanilla and it is easy to modify and doesn't need
96

memory units, therefore, faster to train than LSTM and give as
per performance.
As we compared our results with William Yang's result we
found that CNN is the best among all the results as CNN tend
to be much faster (~5 times faster) than RNN and more
efficient depends on our implementation and because of
Nvidia has historically focused much more on CNN than
RNN, as computer vision mostly employs CNN.
VII. CONCLUSION
In recent years, deception detection in online reviews &
fake news has an important role in business, law enforcement,
national security, political due to the potential impact fake
reviews can have on consumer behavior and purchasing
decisions. Researchers used deep learning with the large
dataset to increase in learning and thus get the best results by
using word embedding for extract features or cues that
distinguish relations between words in syntactic and semantic.
In this paper we cover implementation of RNN technique
models (Vanilla, GRU ) and LSTMs that have been proposed
for the detection of online fake news after we prepare our
LAIR dataset applying to prepare data to word embedding to
get vectors of words then entering this vectors to our deep
learning technique, we found that the results of our
experiments are close but GRU(Gated Recurrent Unit) is the
best because it's solving the problems of Vanilla that popular
of gradient vanishing problem and LSTMs (long short-term
memories) which GRU is easy to modify and doesn't need
memory units , so, we gain faster training than LSTM that
effect in our performance ,but for comparing our results with
William's results we found that CNN‟s(Convolutional Neural
Networks ) is the best from other models due to its speed and
its best results and performance for windows .our future
works will increase this accuracy by merging GRU and
CNN‟s to get the best result .
REFERENCES
[1] Erlich, Aaron, et al. "The double-edged sword of mobilizing citizens
via mobile phone in developing countries." Development Engineering
3, 2018, 34-46.
[2] Krishnamurthy, Gangeshwar, et al. "A deep learning approach for
multimodal deception detection." arXiv preprint arXiv: 1803.00344,
2018. 
[3] Conroy, Niall J., Victoria L. Rubin, and Yimin Chen. "Automatic
deception detection: Methods for finding fake news." Proceedings of
the 78th ASIS&T Annual Meeting: Information Science with Impact:
[4] Pérez-Rosas, Verónica, et al. "Automatic Detection of Fake News."
arXiv preprint arXiv:1708.07104 ,2017.
[5] Allcott, Hunt, and Matthew Gentzkow. "Social media and fake news
in the 2016 election." Journal of Economic Perspectives 31.2 ,2017,
211-36.
[6] Kumar S, Shah N.” False Information on Web and Social Media: A
Survey”. arXiv preprint arXiv: 1804.08559, 2018 Apr 23.
[7] Lopez MM, Kalita J.” Deep Learning applied to NLP”. arXiv preprint
arXiv:1703.03091,2017 Mar 9.
[8] Rubin, Victoria L., Yimin Chen, and Niall J. Conroy. "Deception
detection for news: three types of fakes." Proceedings of the 78th
ASIS&T Annual Meeting: Information Science with Impact:
Research in and for the Community. American Society for
Information Science, 2015.
[9] Wang WY.” " liar, liar pants on fire": A new benchmark dataset for
fake news detection”. arXiv preprint arXiv:1705.00648. 2017, May 1
[10] Verhoeven, Ben, and Walter Daelemans. "CLiPS Stylometry
Investigation (CSI) corpus: a Dutch corpus for the detection of age,
gender, personality, sentiment and deception in text." In LREC 2014-
NINTH INTERNATIONAL CONFERENCE ON LANGUAGE
RESOURCES AND EVALUATION, pp. 3081-3085,2014.
[11] Hagiwara, Masato. "A supervised learning approach to automatic
synonym identification based on distributional features." In
Proceedings of the 46th Annual Meeting of the Association for
Computational Linguistics on Human Language Technologies:
Student Research Workshop, pp. 1-6. Association for Computational
Linguistics, 2008
[12] K Young T, Hazarika D, Poria S, Cambria E. “Recent trends in deep
learning based natural language processing.” arXiv preprint
arXiv:1708.02709. 2017 Aug 9.
[13] Bajaj S. “The Pope Has a New Baby! Fake News Detection Using
Deep Learning”.2017 .
[14] Ruchansky N, Seo S, Liu Y. Csi.” A hybrid deep model for fake news
detection”. InProceedings of the 2017 ACM on Conference on
Information and Knowledge Management ,2017 .Nov 6. pp. 797-806,
ACM.
[15] Lepping, Joachim. "Wiley Interdisciplinary Reviews: Data Mining
and Knowledge Discovery.",2018.
[16] Mikolov T, Sutskever I, Chen K, Corrado GS, Dean J. “Distributed
representations of words and phrases and their compositionality.
InAdvances in neural information processing systems”, 2013 ,pp.
3111-3119.
[17] www.kdnuggets.com.
[18] www.wikipedia .com
97