Journal of Physics: Conference Series

PAPER

A Comprehensive Survey of Spam Profile Detection Methods in Online

Social Networks
To cite this article: R. Krithiga and Dr. E. Ilavarasan 2019 J. Phys.: Conf. Ser. 1362 012111

View the article online for updates and enhancements.

This content was downloaded from IP address 255.255.255.254 on 14/11/2019 at 10:02

International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

A Comprehensive Survey of Spam Profile Detection

Methods in Online Social Networks

R. Krithiga1, Dr. E. Ilavarasan2

1
Assistant Professor, Department of Computer Applications, Perunthalaivar Kamaraj Arts College,
Puducherry
2
Professor, Department of Computer Science & Engineering, Pondicherry Engineering College,
Puducherry
[email protected], [email protected]

Abstract—Social networks have grown into a popular way for internet surfers to interact with friends in
addition to family members, reading news, and also discuss events. Users spend more time on popular
social platforms (e.g., Facebook, Twitter, etc.) storing and sharing their personal information. This fact
together with the prospect of communicating thousands of users fascinates the concentration of malicious
users. They exploit the implicit trust interactions concerning users with the purpose of accomplishing their
malicious objectives, for instance, create malicious links inside the posts/tweets, spread fake news, send
out unsolicited messages to genuine users and so on. In this paper, we reviewed various existing
techniques on spam profile detection in online social networks.
Index Terms—Spam detection, Online Social Networks, Machine learning techniques, Social network
security.

1. Introduction
Recently, Social networking sites (SNSs) have been changed into a crucial and prominent medium of
communication and sharing of knowledge. Ultimately, SNS users are liable for sharing knowledge in
the network as they are content providers in the network structure. The communities which include
families, group of friends, and acquaintances are the primary component in the structure of a
network. In SNSs, users can share information through posting links to their favorite web pages,
files, photos, and videos.
Moreover, the SNS communities’ structure produces a network of credibility in addition to trust
(Lee, 2015 and Singh, et.al, 2017). Facebook and Twitter are the most important SNSs. As per the
report from Statista (Statista report, 2017), the total number of Facebook and Twitter users stood at
1,968 million and 319 million, respectively, as of April 2017. With the increasing number of users,
an enormous amount of diverse knowledge is also being produced every day on these two SNSs
(Sharma, et.al, 2017). Mainly, multimedia knowledge (in a text, audio files, videos, and images
form) is produced, stored, and transferred in vast amounts. The multimedia knowledge posted on
SNSs is often accompanied by user likes, comments, tags, hashtags, and so on.
Generally, spammers on Facebook as well as Twitter pose as legitimate users. Consequently,
recognizing them in addition to distinguishing from legitimate users is a challenging task. In the past,
spammers on these SNSs were usually simple, with a clear appearance that assisted in distinguishing
them from legitimate users. However, spammers can still utilize inexpensive automated approaches
on behalf of acquisition of credibility in addition to trust and making them hard to perceive in the
huge population of SNS users. Detection of a spammer in SNSs is a classification issue in which
legitimate users are well-known from spammers based on their corresponding features. Note,
however, that these classification problems are associated with several unique challenges when
attempting to identify spammers. These challenges includes high dimensionality of features, biased
and less number of training sets, great computational classification difficulties, and public

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

inaccessibility of training sets. To overcome these challenges and achieve spammer classification
accuracy, a number of traditional machine-learning techniques, like Support Vector Machine
(Benevenuto, et.al, 2010 and Zheng, et.al, 2015), Decision Tree (Amleshwaram, et.al, 2013), Jrip
(Ahmed, et.al, 2013), Bayesian Network (Rathore, et.al, 2017), Random Forest (Liu, et.al, 2016), k-
Nearest Neighbors (Herzallah, et.al, 2017), etc. were used in the previous researches. Due to the high
dimensionality of features, however, determining the optimum parameters for parametric supervised
procedures are very slow and challenging (Zheng, et.al, 2016). Hence, existing models suffer from
numerous problems, like poor generalization performance, extended training time, and greater false
positive rates. Additionally, several methods, similar to (Miller, et.al, 2014), utilized biased dataset
enclosing a much smaller number of spam profiles than legitimate ones in addition to offering
imprecise classification outcomes.

2. Literature Review
This section deals with the previous studies related to spam profile detection in Online Social
Networks (OSN). Some of the noteworthy studies include the following.
Aswani, et.al, in 2018 proposed a novel hybrid approach by combining analytics from social media
and bio-inspired computing intended with the view of identifying spam profiles in twitter. The K-
Means algorithm was combined with Levy flight Firefly Algorithm (LFA) along with chaotic maps.
For this study, a total of 18,44,701 tweets were examined from a total of 14,235 Twitter profiles
based on 13 statistically significant attributes from OSN analytics. The outcomes exhibited an
accuracy of around 97.98% by examining the tweets based on those 13 attributes. One of the major
disadvantages observed from the proposed system was the lack of consideration of content and
semantics aspects, as most of current tweets include parody besides usage of non-English
terminologies of millennial language (Aswani, et.al, 2018).
In the same year, Fu, et.al, developed a dynamic metric for measuring the alterations in activities of
users through the influence of progressive evolution patterns of OSN users. This was made
accomplished by combining unsupervised as well as supervised learning with the view of
distinguishing the variances among spammers and legitimate users. A real-world dataset comprising
of a huge sum of users were tested under this methodology. The proposed technique effectively
classified spammers and legitimate users based on their patterns of temporal evolution. Similarly, it
demonstrated an extended level of resemblances in temporal evolution patterns of spammers (Fu,
et.al, 2018).
Similarly, Dutta et.al developed an attribute selection methodology for enhancing the classification
accuracy of spams, which provided better classification results by selecting a smaller subset of
attributes. This attribute assortment algorithm was developed by applying the conceptions of rough
set principles. Experiments were conducted over five different spam classification datasets and were
validated with the performance of the proposed algorithm. The proposed technique selected a much
smaller subset of attributes than the standard techniques with promising classification results than the
other techniques in the literature (Dutta, et.al, 2018).
Eshraqi et.al, in 2015, reviewed various works related to spam detection in social networks. Most of
the reviews in this study concentrated on social networking sites like Twitter and Facebook.
According to the author, the study has been divided into five sections in which the collection of
various data regarding spam profiles in social networks was made as the first phase. The examination
of features of spam user account detection was carried out in the second phase. The third phase
comprised of detection in features of a spam post while the fourth stage was devoted to the
examination of different classification and clustering techniques used for detecting spam. And
finally, the fifth stage summarised the results obtained through all the reviews (Eshraqi, et.al, 2015).
Gao, et.al, in 2012 developed a system for filtering online spams which could be organized in the
OSN platform for examining the messages generated by users. Under this system, the spam
messages rather than being inspected distinctly were reconstructed to their respective campaigns for
classification. Almost 187 million wall posts from Facebook and 17 million tweets from Twitter

2
International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

were collected and examined using this proposed structure. As a result of advanced examination
through the proposed system, a message alerting “spam” was dropped before the spam message to
assist the users.(Gao, et.al, 2012).
A novel and more sturdy attributes for identifying Twitter spammers were designed by Yang et.al, in
2013. This was based on an in-depth analysis strategy related to evasion which was generally utilized
by spammers in Twitter. Under this study, at the initial stage, experimental analyses were made to
have an in depth understanding of evasion tactics employed by spammers. Subsequently, 24
attributes for detection process were analysed for their robustness. It can be noted from the study that
even under a low false positive rate, the detection rate of this proposed system was considerably
higher than the other existing spam indicator techniques under four different prevalent machine
learning classifiers (Yang, et.al, 2013).
In 2015, a performance evaluation was made by Chen, et.al, over three distinctive aspects of data,
feature as well as a model. For this evaluation, a dataset comprising of almost 600 million public
tweets was generated by means of a commercial URL-based security platform. Here, for detecting a
tweet with spam message, 12 lightweight features were extracted from the tweet. As an extension,
this detection was then transformed into a binary classification problem which could be solved using
conventional machine learning algorithms. The performance evaluation was made over the attributes
of spam to non-spam ratio, feature discretization, the training data size, data sampling, procedures
related with machine learning and data regarding time (Chen, et.al, 2015).
In the year 2017, Liu, et.al suggested a fuzzy-based oversampling method that offered artificial data
samples. These samples were obtained from limited observed samples depending on their idea of
classification using fuzzy based information. Moreover, an ensemble technique was developed which
examined certain classifiers based on their imbalanced data in three phases. It could be concluded
from the study that by means of imbalanced class distribution, the rate of spam detection can greatly
be influenced (Liu, et.al, 2017).
A set of characteristic attributes from Twitter, concerning network, activity, user, as well as tweet
content were utilized in developing a supervised machine learning solution for recognizing
cyberbullying on Twitter. This was suggested by Al-garadi et.al, in 2016 which resulted in an f-
measure of 0.936 with an area under the receiver-operating characteristic curve of 0.943. These
outcomes suggested that the model based on these features offered a reasonable solution for
identifying Cyberbullying in online communication backgrounds. Lastly, the outcome acquired with
the proposed features was proven the best by comparing with results from two baseline features for
their performance (Al-garadi, et.al, 2016).
A novel methodology related to Markov Clustering (MCL) was presented by Ahmed & Abulaish in
2012 for determining the spam profiles over OSNs. A dataset comprising of Facebook profiles (both
real and spam profiles) was utilized for the experiment. A weighed graph was applied which helped
in modeling the social networks in which the profiles were characterized as nodes and their
intersections as edges. Also, in case of a cluster comprising both real and spam profiles, the idea of
majority voting was established. The outcomes have revealed that the introduction of majority voting
not only reduces the sum of clusters but also increases the performance values of FP and FB measures
(Ahmed & Abulaish, 2012).
An effective spammer detection outline which distinguished the legitimate users of Facebook from
the spammers was proposed by Rathore, et.al, in 2018. In accordance with recent activities of
Facebook, the outline presented the urgency of setting out a new feature set for detecting spams. A
baseline dataset was used from Facebook which involved 300 spammers in addition to 700
legitimate user profiles. The baseline dataset encompassed feature set for an individual profile, which
were extracted through a novel dataset construction mechanism. Furthermore, an intelligent decision
support scheme was employed to differentiate spammers from genuine users. The evaluation
outcomes proved that the suggested outline was precise and effective to deliver excellent
performance with a maximum accuracy of 0.984 and Mathew correlation coefficient of 0.977
(Rathore, et.al, 2018).

3
International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

A study related to spot the fake profiles created by people in contrast to those created by bots was
reported by Van, et.al, in 2018. Initially, a series of early works regarding spam detection techniques
were studied for any resemblance with their report. A number of supervised machine learning
models were used with features exclusively engineered for spam profile detection without trusting
behavioural data. The results of the study witnessed an F1 score of 49.75% upon detection of fake
accounts (Van, et.al, 2018).
An integrated technique with added categories of features like metadata, interaction based as well as
content-based attributes to previously existing community-based features for recognizing auto
spammers was discussed by Fazil, et.al, in 2018. The novelty of this technique is that the users were
categorized based on their exchanges with their followers as ignoring follower’s contents and
metadata features were literally difficult. The experiment was made over a real-world dataset
comprising of legitimate users as well as the spammers whose profiles were characterized through 19
predefined attributes along with 6 newly defined and 2 redefined features. The results showed that
the classifications based on interaction as well as community proved to be effective while metadata
classification was the least effective (Fazil, et.al, 2018).
With the view of recognizing spammers in Twitter, a hybrid machine learning model was proposed
by hybridizing an existing meta-heuristic technique called Whale Optimization Algorithm (WOA)
with Support Vector Machines (SVM). The idea behind the utilization of WOA in this hybrid
technique was to enhance the parameters of SVM along with the task of selecting suitable features
for spam recognition. Arabic, English, Spanish, as well as Korean lingual contexts, were recognized
under this technique. The results have shown a considerable efficiency over other previously
developed spam profile detection techniques. (Ala’M, et.al, 2018).
An alternative ontology-based methodology was proposed to detect spam tweets exclusively by
means of content analysis. This was carried out as the accuracy of classification was much lower
than the expected. The proposed ontology disabled the dependency on private as well as user-
relationship information which most of the existing spam detection procedures use. Investigational
outcomes demonstrated that the proposed method outperformed existing message to message spam
detection methods by approximately 200% (Halawi, et.al, 2018).
A study based on spammers account observed over the OSNs was conducted using behavioural
features by Almaatouq, et.al, in 2016. This study was carried out over a huge sum of 100 million
tweets from Twitter collected over a span of a month. It identified two behaviourally dissimilar
groups of spammers while they utilize different approaches of spamming. Similarly, it illustrated the
way in which the legitimate, as well as the spammers in these groups, validate each individual
property along with their patterns of social interaction. Finally, these groups were checked for their
content attributes, social communication, as well as features related to profile for detecting spam
messages (Almaatouq, 2016).
An innovative methodology was proposed that uses Open-source Intelligence (OSINT) along with
machine learning techniques for localizing and outlining the movement of a social network user.
This structure identified the location of users through their posts thereby raising the privacy issues
involved. The outcomes have revealed an estimated accuracy of 77.72% stating that the users can be
geo-located through this proposed technique without the use of GPS/GNSS related data (Pellet, et.al,
2019).
A real-time anti-phishing structure comprising of seven distinct procedures of classification along
with Natural Language Processing (NLP) based attributes was suggested by Sahingoz, et.al, in 2019.
Based on the following properties of language independence, usage of a large amount of phishing as
well as authentic data, implementations in real-time, identification of new websites, independent
from third-party services and usage of feature-rich classifiers, studies were carried out for
recognizing their distinctive features. The proposed system was then tested with a novel dataset
comprising of the above mentioned properties. As per the investigational and comparative outcomes
from the implemented classification procedures, Random Forest algorithm with only NLP based
features provides the improved performance with the 97.98% accuracy rate for recognition of
phishing URLs (Sahingoz, et.al, 2019).

4
International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

Based on the level of focused interest patterns of users, an advanced spammer detection system was
proposed by Alghamdi, et.al, in 2018. For recognizing the variation in interests of users, quantitative
techniques were proposed and defined for the type of interest a user possess i.e. either focussed type
or diverse-interest type. Based on the interest level, the users were recognized and a framework was
developed by incorporating both unsupervised as well as supervised learning for categorizing
between legitimate and spam users (Alghamdi, et.al, 2018).
Crowd-retweeting of spam in Sina Weibo, the counterpart of Twitter in China was examined by Liu
et al., in 2018. The characteristics of crowd retweeting spammers were carefully analyzed
concerning their profile features, social relationships, and retweeting behaviours. The authors found
that even though these spammers were probably as close as to legitimate users, the process of basic
social connections of crowd retweeting campaigns were totally different from those of prevailing
spam campaigns. This was recognized through the unique features of retweets that were spread in the
cascade. Based on these outcomes, retweeting-aware link-based ranking algorithms were proposed to
conclude more suspicious accounts by preferring recognized spammers as seeds. The evaluation
outcomes presented that the procedures were more efficient compared to other link-based strategies
(Liu, et.al, 2018).
In 2018, a structured classification approach for spammer detection problem on OSNs was suggested
by El-Mawass et al. This approach influenced the similarities amidst a range of users for transmitting
the beliefs about their labels. A Markov Random Field model was generated in the form of a
similarity graph by associating various analogous profiles based on their shared applications. It not
only allowed the detection system to be more sustainable but utilized in designing adaptive
classifiers. It could be concluded from the study that the detection accuracy of any technique can be
influenced through similarities between user profiles (El-Mawass, et.al, 2018).
A qualitative examination was made over the advantages and disadvantages of various previously
developed spam detection techniques. Every detection technique was comprehensively reviewed and
as a result, a discussion comprising of several gaps prevalent in existing approaches and the
corresponding actions needed to address them were clearly stated (Kaur, et.al, 2018).
The automatic discovery of cyber security related social network based on three diverse sets of a
feature in addition to three unique machine learning approaches was proposed by Aslan et al., in
2018. Investigational outcomes presented a favorable performance with an extraordinary accuracy of
95%. The uppermost score (over 97%) was attained by combining the random forests technique with
behavioural features. Maintaining a list of cyber security linked accounts manually requires domain-
specific knowledge and consumes human efforts. As an outcome of this study, it was stated that a list
can automatically be maintained which would help in complicated analysis like cyber security
attacks, human behaviours of cyber criminals and cybersecurity experts by automatically checking
the accounts in the list (Ashlan, et.al, 2018).
A set of features were adopted for recognizing spammers on Twitter along with some additional
features for enhancing the performance of classifiers by Hanif, et.al, in 2018. WEKA and
RapidMiner were used for estimating the performance of four machine learning algorithms namely
Random forest (RF), Support vector machine (SVM), K nearest neighbor (KNN), and Multilayer
Perceptron (MLP). The experiment showed that SVM, KNN, and MLP on WEKA outperformed
those techniques on RapidMiner. However, in the case of RF, RapidMiner attained greater accuracy
compared to RF on WEKA. Based on the 32 features in the dataset, MLP in addition to RF on both
WEKA and RapidMiner outperformed compared to other classifiers with an accuracy of 95.42% and
95.44% respectively (Hanif, et.al, 2018).
In 2018, machine learning based malicious account detection scheme called DeepScan for Location-
Based Social Networks (LBSNs) was proposed by Gong et al. By separating every user’s activity
information into many continuous time intervals, the users’ activities were measured. Through deep
learning technologies, DeepScan makes application of time series features, which were more
discriminative and informative, compared to conventional features, specifically statistical or
demographic features. Through the real data collected from Dianping, the authors found that the
DeepScan achieved an outstanding performance with an F1-score of 0.964 (Gong, et.al, 2018).

5
International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

A novel dataset based on various features collected from numerous latest research works was created
for developing a more robust as well as accurate spammer detection model. This dataset was then
tested with various standard classifiers for their prediction performance. Furthermore, an extended
analysis was made for identifying the features that possess a higher impact of accuracy in detecting
spam. For analysis, three different methodologies were chosen: change of mean square error (CoM),
information gain (IG) and Relief-F method and compared for their performance effectiveness. The
attributes such as the reputation of a profile, average words in a tweet, average mention per tweet,
the active period of an account and average time within posts contributes more in detecting
spammers in OSNs (Herzallah, et.al, 2018)
A Bagging ELM-based spammer detection framework for SNSs was proposed by Rathore et al., in
2018. The proposed framework had three major contributions in the area. Initially, it identified
account and object-specific features to enable spammer detection in SNSs. Next, it built a novel
dataset of the two most popular SNSs, i.e., Twitter and Facebook. Lastly, it presented a Bagging
ELM classifier and applied this classifier to the dataset that was constructed from Twitter and
Facebook. The experiments and assessment with other classifiers showed that the proposed structure
was capable of attaining improved generalization performance than other frameworks. The proposed
system attained an average rate of 99.01 % accuracy for the Twitter dataset and 99.02 % for the
Facebook dataset while requiring shorter training time of 1.17s and 1.10s, individually (Rathore,
et.al, 2018).
Perez et al., (2018) considered Twitter as a case study to measure the distinctiveness of the
relationship concerning metadata and user uniqueness and comprehend the effects of potential
obfuscation approaches. More precisely, atomic fields in the metadata were examined and
scientifically integrated with a view of classifying new tweets belonging to an account through
various machine learning algorithms. The authors demonstrated that, with the application of
supervised learning procedure, they were capable of recognizing any user from a pool of of 10,000
profiles with around 96.7% accuracy. (Perez, et.al, 2018).
A graph-based technique for identifying tourist movement patterns from Twitter information was
presented by Hu et al. Initially, tweets collected using geo-tags were gutted to filter those not issued
by tourists. Next, a DBSCAN-based clustering technique was modified to construct tourist graphs
containing the tourist attraction vertices and edges. Third, network analytical approaches were used
to identify tourist movement patterns, including popular and centric attractions, and popular tour
routes. The New York city in US was considered to validate the proposed methodology utility. The
detected tourist movement patterns support business in addition to government activities such as tour
planning, transportation, and enlargement of both shopping and accommodation centers (Hu, et.al,
2018).
Wani et al recommended a model that can be opted by the OSN service providers to alert their users
with a list of suspicious connection (links) from their corresponding friend lists so that users can
themselves authenticate the recommended links and filter their friend list based on the alerts. Data
from 1000 interconnected Facebook profiles were collected for designing the classifier proposed
using mutual clustering coefficient metric (Wani, et.al, 2018).
Sohrabi et al. recommended a novel technique regarding the recognition of spam comments on
Facebook. In this study, an online spam filtering scheme was considered through revising the posts in
addition to comments and reviewing their features. The suggested filtering system was capable of
exploiting different techniques and optimization procedures such as simulated annealing, PSO, ACO,
and differential evolution to perceive and remove the impish contents and avoids issuing spam
comments to offer a secure environment for users of the standard social network. Moreover,
supervised machine learning techniques, clustering methods, and decision trees were used for the
suggested filtering scheme (Sohrabi, et.al, 2018).
In 2010, an exploration regarding the limits of classifying social spam profiles on MySpace was
presented by Irani, et.al. The study focussed on zero-minute fields or commonly known as static
fields which were presented during the creation of a profile in order to try and determine whether this
sort of technique would be feasible in preventing spammers. The results displayed a standard C4.5

6
International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

decision tree performs better with an AUC of 0.994 with an average of approximately 24 false-
positives and 70 false-negatives (Irani, et.al, 2010).
A feature-based technique was developed along with a supervised learning method for detecting
spam posts from Instagram. K-fold cross-validation was utilized for determining the optimum pair of
supervised learning model as well as the parameters for the model. Many profiles from Instagram
were collected and for marking media posts quickly a two-pass clustering method (i.e. Minhash
clustering and K-medoids clustering) was utilized. This was primarily formed in order to group the
near-duplicate posts into a similar form of clusters. It can be witnessed that the accuracy of this
proposed system was around 96.27% (Zhang & Sun, 2017).
Adikari & Dutta studied the minimal set of profile data that were required for determining the fake
profiles in LinkedIn and identified the appropriate data mining technique for carrying out such a
task. In this study, two predominant techniques of data mining namely RBF kernel and Polynomial
kernel were compared for their accuracy and a better technique was chosen among these. Although
the former technique remains as the most commonly utilized method in data mining context, the
latter provides higher accuracy. The accuracy of the proposed polynomial kernel was found to be
84.04 % (Adikari & Dutta, 2014).
A new set of features through EdgeRank algorithm was proposed for detecting the video spammers
on YouTube. This study comprised of experiments with nine classifiers of various learning, decision
tree, function-based and Bayesian techniques. Through the study, it was witnessed that a maximum
average of 98 percent was observed over Bayes Network and Naïve Bayesian network while the
minimum was found to be 94 % percent with LibLINEAR technique (Yusof & Sadoon, 2017).

Table 1: Comparison of Existing Techniques

Author Advantages Limitations Social Metrics Size of
S.No Networks Dataset
examined
1. Gao et Extraordinary Although this Facebook High 187
al., accuracy, latency system detects and accuracy, low million
(2012) at lower levels fake accounts in Twitter. latency and wall posts
and high Facebook and high from
throughput were Twitter, it cannot throughput. Facebook
some of the be greatly used as and 17
advantages of this they are not million
system. Also, they generic. They tweets
incur low cost for focus only on from
maintenance after detecting twitter.
deployment. accounts that
spread URLs
through
messages.
2. Yang et While keeping an It uses a crawled Twitter High detection 20000
al., even lower false dataset which rate, low false twitter
(2013) positive rate, the may still possess positive rate. accounts
detection rate a sampling bias.
using the new This deteriorates
proposed feature the accuracy of
set is significantly spam profile
higher than that of detection
existing features. technique.

7
International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

3. Chen et In the evaluation, Performance Twitter Detection 600

al., it was found that decreases because ability, feature million
(2015) the classifiers’ of the distribution discretization, tweets
ability to detect of features spam
Twitter spam changes of later detection rate.
reduced when in a days’ dataset,
near real-world whereas the
scenario since the distribution of
imbalanced data training dataset
brings bias. The stays the same.
proposed method This problem will
overcome this exist in streaming
scenario spam tweets
effectively. detection, as the
new tweets are
coming in the
forms of streams,
but the training
dataset is not
updated.
4. Liu et Improves the The synthetic Twitter Spam 600
al., spam detection data generation detection million
(2017) performance on scheme to performance tweets
imbalanced incorporate
Twitter datasets correlations
with a range of among features
imbalance was not done.
degrees
5. Al- The proposed Not utilized the Twitter Precision, F- 2.5
garadi model can be data from other measure, million
et al., used by the social media to recall and geo-
(2018) organization’s investigate cyber AUC. tagged
members as well bullying tweets
as non- behaviours
government
organizations
(NGOs),
including crime-
prevention
foundations,
social chamber
organizations,
psychiatric
associations,
policymakers, and
enforcement
bodies.
6. Jantan FFNN-EBAT has Not verified the (e-mails) Mean, median 8467 e-
et al., better quality performance of and standard mail
(2017) results than the the EBAT deviation of instances
other training algorithm with Mean Square
algorithms other neural Error (MSE).
network types,

8
International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

and not
investigated its
effectiveness with
other spam email
datasets.
7. Rathore More accurate Specific to Facebook Efficiency and 1000
et al., and efficient to facebook and accuracy. Facebook
(2018) deliver excellent cannot be applied user
performance. Fast to other SNSs. profiles.
response and high
accuracy rate for
spammer
detection on
Facebook.

8. Ala'M This model can be The proposed Twitter Accuracy, 800

et al., helpful in system detects 4 Precision, F- instances
(2018) designing more major languages measure and
accurate and and a multilingual AUC
insightful spam source. Areas
detection models such as hashtag
for online social usages in English,
networks. The tweets via mobile
model can also be in Arabic, number
helpful in of tweets per day
identifying the in Korean etc.,
most influencing were not
features. considered

9. Pellet et Localization of The proposed Twitter, Prediction 5447

al., user was made technique acts Instagram accuracy. instances
(2019) possible through only on three and
the developed major OSNs Facebook.
social graph namely,
technique. Facebook,
Twitter and
Instagram and if
the candidate
does not have
account in these
networks, then
the localization
would not be
possible through
this technique.
10. Singh The model can be The framework Twitter. Performance, -
& Batra successfully was tested for Precision, F-
(2018) deployed for tweets generated measure and
spam detection in by Twitter alone recall.
massive data sets which demands
further
enhancement to

9
 International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

include OSNs
too.

3. Conclusion
Due to the increasing popularity and intensive use of social networks such as Twitter, Facebook,
LinkedIn, etc. the number of spammers are rapidly growing. This has resulted in the development of
several spam detection techniques. From the papers reviewed, it can be concluded that most of the
work have been developed using classification approaches like SVM, Decision Tree, Naive
Bayesian, and Random Forest, KNN. The effectiveness of methods has been determined based on
the metrics such as detection accuracy, efficiency, false prediction rate, F-Measure, Precision and
recall. Detection rate is measured from user-based or content based features or a combination of both
and network level features. It is to be noted that spam detection methods have been proposed
considering only any one of the of social network type which is not effective in the current scenario
as an individual has accounts across various social networks. There is a huge scope of improvements
in terms of unified system for spam identification, introduction and exploration of new features
which would account the activities of users at various time frames and autonomous system that
updates itself at periodic intervals so as to defend against the evolving and invading spammers.

4. REFERENCES
[1]. Adikari, S., & Dutta, K. (2014), “Identifying Fake Profiles in LinkedIn”, In PACIS (p. 278).
[2]. Ahmed, F., & Abulaish, M. (2012, June), “An mcl-based approach for spam profile detection in
online social networks”. In Trust, Security and Privacy in Computing and Communications
(TrustCom), 2012 IEEE 11th International Conference on (pp. 602-608), IEEE.
[3]. Ahmed, F., & Abulaish, M. (2013), “A generic statistical approach for spam detection in Online
Social Networks”, Computer Communications, 36(10-11), 1120-1129.
[4]. Ala’M, A. Z., Faris, H., & Hassonah, M. A. (2018), “Evolving Support Vector Machines using
Whale Optimization Algorithm for spam profiles detection on online social networks in different
lingual contexts”, Knowledge-Based Systems, 153, 91-104.
[5]. Al-garadi, M. A., Varathan, K. D., & Ravana, S. D. (2016), “Cybercrime detection in online
communications: The experimental case of cyberbullying detection in the Twitter
network”, Computers in Human Behavior, 63, 433-443.
[6]. Alghamdi, B., Xu, Y., & Watson, J. (2018, November), “A Hybrid Approach for Detecting
Spammers in Online Social Networks”, In International Conference on Web Information
Systems Engineering (pp. 189-198), Springer, Cham.
[7]. Almaatouq, A., Shmueli, E., Nouh, M., Alabdulkareem, A., Singh, V. K., Alsaleh, M., &
Pentland, A. S. (2016), “If It Looks Like a Spammer and Behaves Like a Spammer, It Must Be a
Spammer. Analysis and Detection of Microblogging Spam Accounts”, International Journal of
Information Security 15.5 (2016): 475–491.
[8]. Amleshwaram, A. A., Reddy, N., Yadav, S., Gu, G., & Yang, C. (2013, January), “Cats:
Characterizing automation of twitter spammers”, In Communication Systems and Networks
(COMSNETS), 2013 Fifth International Conference on (pp. 1-10). IEEE.
[9]. Aslan, Ç. B., Sağlam, R. B., & Li, S. (2018), “Automatic Detection of Cyber Security Related
Accounts on Online Social Networks: Twitter as an example”, In: 9th International Conference
on Social Media & Society (SM&Society 2017), 18-20 July 2018, Copenhagen, Denmark.
(doi:https://doi.org/10.1145/3217804.3217919)
[10]. Aswani, R., Kar, A. K., & Ilavarasan, P. V. (2018), “Detection of spammers in twitter marketing:
A hybrid approach using social media analytics and bio-inspired computing”, Information
Systems Frontiers, 20(3), 515-530.

10
 International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

[11]. Benevenuto, F., Magno, G., Rodrigues, T., & Almeida, V. (2010, July), “Detecting spammers on
twitter” In Collaboration, electronic messaging, anti-abuse and spam conference (CEAS) (Vol. 6,
No. 2010, p. 12).
[12]. Chen, C., Zhang, J., Xie, Y., Xiang, Y., Zhou, W., Hassan, M. M., & Alrubaian, M. (2015), “A
performance evaluation of machine learning-based streaming spam tweets detection”, IEEE
Transactions on Computational social systems, 2(3), 65-76.
[13]. Dutta, S., Ghatak, S., Dey, R., Das, A. K., & Ghosh, S. (2018), “Attribute selection for
improving spam classification in online social networks: a rough set theory-based
approach”, Social Network Analysis and Mining, 8(1), 7.
[14]. El-Mawass, N., Honeine, P., & Vercouter, L. (2018, July), “Supervised Classification of Social
Spammers using a Similarity-based Markov Random Field Approach”, In Proceedings of the 5th
Multidisciplinary International Social Networks Conference (p. 14). ACM.
[15]. Eshraqi, N., Jalali, M., & Moattar, M. H. (2015, November), “Spam detection in social networks:
A review”, In Technology, Communication and Knowledge (ICTCK), 2015 International
Congress on (pp. 148-152). IEEE.
[16]. Fazil, M., & Abulaish, M. (2018), “A Hybrid Approach for Detecting Automated Spammers in
Twitter”, IEEE Transactions on Information Forensics and Security, 13(11), 2707-2719.
[17]. Fu, Q., Feng, B., Guo, D., & Li, Q. (2018), “Combating the evolving spammers in online social
networks”, Computers & Security, 72, 60-73.
[18]. Gao, H., Chen, Y., Lee, K., Palsetia, D., & Choudhary, A. N. (2012, February), “Towards Online
Spam Filtering in Social Networks”, In NDSS (Vol. 12, pp. 1-16).
[19]. Gong, Q., Chen, Y., He, X., Zhuang, Z., Wang, T., Huang, H., & Fu, X. (2018), “DeepScan:
Exploiting Deep Learning for Malicious Account Detection in Location-Based Social
Networks”, IEEE Communications Magazine, Feature Topic on Mobile Big Data for Urban
Analytics, 56(1).
[20]. Halawi, B., Mourad, A., Otrok, H., & Damiani, E. (2018), “Few are as Good as Many: An
Ontology-Based Tweet Spam Detection Approach”, IEEE Access.
[21]. Hanif, M. H. M., Adewole, K. S., Anuar, N. B., & Kamsin, A. (2018), “Performance Evaluation
of Machine Learning Algorithms for Spam Profile Detection on Twitter Using WEKA and
RapidMiner”, Advanced Science Letters, 24(2), 1043-1046.
[22]. Herzallah, W., Faris, H., & Adwan, O. (2018), “Feature engineering for detecting spammers on
Twitter: Modelling and analysis”, Journal of Information Science, 44(2), 230-247.
[23]. Hu, F., Li, Z., Yang, C., & Jiang, Y. (2018), “A graph-based approach to detecting tourist
movement patterns using social media data”, Cartography and Geographic Information Science,
1-15.
[24]. Irani, D., Webb, S., & Pu, C. (2010, May), “Study of Static Classification of Social Spam
Profiles in MySpace” In Icwsm.
[25]. Jantan, A., Ghanem, W. A., & Ghaleb, S. A. (2017), “Using Modified Bat Algorithm To Train
Neural Networks For Spam Detection”, Journal of Theoretical & Applied Information
Technology, 95(24).
[26]. Kaur, R., Singh, S., & Kumar, H. (2018), “Rise of spam and compromised accounts in online
social networks: A state-of-the-art review of different combating approaches”, Journal of
Network and Computer Applications.
[27]. Lee, D. H. (2015), “Personalizing Information Using Users' Online Social Networks: A Case
Study of CiteULike”, JIPS, 11(1), 1-21.

11
 International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

[28]. Liu, B., Ni, Z., Luo, J., Cao, J., Ni, X., Liu, B., & Fu, X. (2018), “Analysis and defense against
crowd-retweeting based spam in social networks”, World Wide Web, 1-23.
[29]. Liu, L., Lu, Y., Luo, Y., Zhang, R., Itti, L., & Lu, J. (2016), “Detecting Smart Spammers On
Social Network: A Topic Model Approach”, arXiv preprint arXiv:1604.08504.
[30]. Liu, S., Wang, Y., Zhang, J., Chen, C., & Xiang, Y. (2017), “Addressing the class imbalance
problem in Twitter spam detection using ensemble learning”, Computers & Security, 69, 35-49.
[31]. Miller, Z., Dickinson, B., Deitrick, W., Hu, W., & Wang, A. H. (2014), “Twitter spammer
detection using data stream clustering”, Information Sciences, 260, 64-73.
[32]. Mv Ngo Tien HoA,High Speed And Reliable Double Edge Triggered D- Flip-Flop For Memory
Applications”,Journal of VLSI Circuits And Systems, 1 (01), 13-17,2019
[33]. MN BORHAN,”Design Of The High Speed And Reliable Source Coupled Logic
Multiplexer”,Journal of VLSI Circuits And Systems 1 (01), 18-22,2019
[34]. Metadata Information”, arXiv preprint arXiv:1803.10133.
[35]. Rathore, S., Loia, V., & Park, J. H. (2018), “SpamSpotter: an efficient spammer detection
framework based on intelligent decision support system on facebook”, Applied Soft
Computing, 67, 920-932.
[36]. Rathore, S., Sangaiah, A. K., & Park, J. H. (2018), “A novel framework for internet of
knowledge protection in social networking services”, Journal of computational science, 26, 55-
65.
[37]. Sahingoz, O. K., Buber, E., Demir, O., & Diri, B. (2019), “Machine learning based phishing
detection from URLs”, Expert Systems with Applications, 117, 345-357.
[38]. Sharma, P. K., Rathore, S., & Park, J. H. (2017), “Multilevel learning based modeling for link
prediction and users’ consumption preference in Online Social Networks”, Future Generation
Computer Systems.
[39]. Singh, A., & Batra, S. (2018), “Ensemble-based spam detection in social IoT using probabilistic
data structures”, Future Generation Computer Systems, 81, 359-371.
[40]. Sohrabi, M. K., & Karimi, F. (2018), “A feature selection approach to detect spam in the
Facebook social network”, Arabian Journal for Science and Engineering, 43(2), 949-958.
[41]. Statista, Most famous social network sites worldwide as of April 2017, ranked by a number of
active users (in millions).https://www.statista.com/statistics/272014/global-social-networks-
ranked-by-number-of-users/ (accessed 13 September 2017).
[42]. Van Der Walt, E., & Eloff, J. (2018), “Using Machine Learning to Detect Fake Identities: Bots
vs. Humans”, IEEE Access, 6, 6540-6549.
[43]. Wani, M. A., & Jabin, S. (2018), “Mutual Clustering Coefficient-based Suspicious-link
Detection approach for Online Social Networks”, arXiv preprint arXiv:1805.00537.
[44]. Yang, C., Harkreader, R., & Gu, G. (2013), “Empirical evaluation and new design for fighting
evolving twitter spammers”, IEEE Transactions on Information Forensics and Security, 8(8),
1280-1293.
[45]. Yusof, Y., & Sadoon, O. H. (2017), “Detecting video spammers in youtube social media”, in
Zulikha, J. & N. H. Zakaria (Eds.), Proceedings of the 6th International Conference of
Computing & Informatics (pp 228-234). Sintok: School of Computing.
[46]. Zhang, W., & Sun, H. M. (2017, January). “Instagram Spam Detection”. In Dependable
Computing (PRDC), 2017 IEEE 22nd Pacific Rim International Symposium on (pp. 227-228).
IEEE.

12
 International Conference on Physics and Photonics Processes in Nano Sciences IOP Publishing
Journal of Physics: Conference Series 1362 (2019) 012111 doi:10.1088/1742-6596/1362/1/012111

[47]. Zheng, X., Zeng, Z., Chen, Z., Yu, Y., & Rong, C. (2015), “Detecting spammers on social
networks”, Neurocomputing, 159, 27-34.
[48]. Zheng, X., Zhang, X., Yu, Y., Kechadi, T., & Rong, C. (2016), “ELM-based spammer detection
in social networks”, The Journal of Supercomputing, 72(8), 2991-3005
[49]. MN BORHAN,”Design Of The High Speed And Reliable Source Coupled Logic
Multiplexer”,Journal of VLSI Circuits And Systems 1 (01), 18-22,2019
[50]. Mv Ngo Tien HoA,High Speed And Reliable Double Edge Triggered D- Flip-Flop For Memory
Applications”,Journal of VLSI Circuits And Systems, 1 (01), 13-17,2019

Author Biography

First Author

Mrs. R. Krithiga, currently works as an Assistant Professor, Department of Computer Applications at

Perunthalaivar Kamarajar Arts College, Madagadipet, Puducherry. She completed her Master of
Computer Science from Pondicherry University. She is also a research scholar at Pondicherry
Engineering College. Her area of interests includes Evolutionary algorithms, Data mining and
Intelligent systems.
SecondAuthor

Dr. E. Ilavarasan, is a Professor of Department of Computer Science & Engineering at Pondicherry

Engineering College, Puducherry. He has an experience of more than 25 years in the teaching field.
He is expertized in Web service computing.
.

13