IEEE Consumer Electronics Magazine - January/February 2020 - 86

Special Section on Cognitive Science and AI for Human Cognition and Communication

Table 2. Performance analysis of the cognitive
identification framework.
Classifier

Accuracy

Detection rate

FPR

B-ELM

96.1%

95.3%

0.82%

ELM

90.4%

90.2%

3.45%

SVM

89.2%

88.1%

5.67%

SNS user relies to recognize other SNS users,
including profile description, profile picture, profile age, etc. Psychological traits surpass at delivering moderate differentiability and tenacity. The
characteristics of user's profiles have been analyzed on Twitter to differentiate human accounts
from the socialbot that are managed using automated program.11
Activity traits are found to be the second class
of cognitive traits, which involve the analysis of
behaviors of SNS users. Our study analyzed that
the malicious users depict different behavior from
ordinary users in terms of content generation
patterns. Thus, extracting the patterns of user's
shared content could help differentiate malicious
users from ordinary ones. In this class of traits,
many types of features related to shared content are typically used to characterize user's activity including textual patterns, content posting
behaviors. Textual patterns involve word length of
content, message similarity, message length, language model, and N-gram to detect the malicious
activity in SNSs. Content posting behaviors describe the posting rate of the non-natural objects,
such as URLs, likes, tags, comments, hashtags,
photos/videos, retweets, mentions, and repost.8
The last type of cognitive traits is social
traits, which describe the social awareness of
SNS users in the context of discriminative network characteristics, such as community membership, degree density, clustering coefficient
centrality measures, and degree distribution. It
has been demonstrated that a user's social relationship including friends, followers, followings,
follower/friend ratio, and user reputation is commonly exploited to verify his/her behaviors.8,12

EXPERIMENTAL ANALYSIS
We performed an experiential evaluation of the
real-time data collected from Twitter, which has
been actively involved in CE marketing in recent

Table 3. Existing research works, their shortcomings,
and a comparison with our framework.
Existing work

Research gap

Von Ahn et al.4

Automated
programs such
as CAPTCHA
can be easily
bypassed using
a malicious
crowdsourcing
service.

Our framework
The cognitive
traits support
prevention
against
malicious
crowdsourcing
services.

Albladi et al.5,
Mulamba et al.6

To implement
social graphs,
it is difficult to
find mutual
trust among
infrequent
users who are
hardly active
or who interact
less with other
users.

Rather than
exploiting
mutual trust,
the framework
uses the GraphX
module of
Apache Sparks
to process the
user's social
connections

Zhou et al.,7
Rathore et al.8

Ineffective in
identifying
adaptive
malicious users
on SNSs

The near-realtime data
analysis
supports the
detection of
adaptive
malicious users.

years. The Twitter streaming API 13 was used to
collect data from Twitter. Based on a list of search
keywords related to various topics (i.e., electronic
products, CE, and life sciences), we collected
45,650 unique tweets created by 2,350 different
users. The data were further supplied to Apache
Spark-2.2.1, which extracts the appropriate cognitive traits of each user and returns the psychological, activity, and social ratings of each user.
We used an intelligent learning algorithm
called Bagging ELM, which has been proposed in
our recent work Rathore et al.12 We also
employed a single ELM and a support vector
machine (SVM) as traditional classifiers to demonstrate the validity of our results over previous
studies. We used three metrics to measure the
performance of our framework: accuracy, detection rate, and false positive rate (FPR).12 The
results are shown in Table 2. Bagging ELM (BELM) achieved a low FPR, high accuracy value,
and a high detection rate. These results demonstrate that our framework can effectively detect
malicious users in SNSs. B-ELM achieved better
performance results than single ELM and SVM.

IEEE Consumer Electronics Magazine

IEEE Consumer Electronics Magazine - January/February 2020

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2020