IEEE Consumer Electronics Magazine - January/February 2024 - 50

Special Section on Security and Privacy-Aware Emerging Computing
Overview
Social media includes a massive amount of data
that help recognize the potential risks in the cryptocurrency
market. Users of social media platforms
interact with each other by liking, commenting,
share actions on posts and these actions can be
analyzed to obtain deep insights into cryptocurrency
risks. Moreover, a linkage between the variation
of cryptocurrency prices and the activities on
social media platforms is driven by empirical evidence.
The identification phase includes identifying
the data related to cryptocurrencies in social
media platforms and then collecting, cleaning,
store them for the forthcoming processes. In the
risk assessment phase, we will evaluate the risks of
cryptocurrencies from the interactions of social
media users. These interactions can be shown in
the number of likes, comments, shares on posts,
and the number of hashtags. So, behavioral features
that would extract from the dataset are the
tool for evaluating risks. It will tell us how, where,
how much, and how long cryptocurrencies users
are generally revealed to a potential risk. This
phase includes three steps namely, Categorization,
feature extraction, and feature selection. In the risk
control stage, we centralize on risks control of cryptocurrencies
by using many methodologies such as
text mining, social network analysis (SNA), sentiment
analysis, and content analysis. The insights
that output from the mining process are visualized
to the decision makers by many techniques. In the
monitoring phase, illustrated perceptions in the
previous step help the cryptocurrency investors
and financial organizations with valuable feedback
about the current risks and how to minimize them.
After the analysis, a possible decision is expected
to be taken by the investors. So, these decisions
will affect the cryptocurrency trading positively or
negatively. Hence, the risk management process is
frequent to recognize and reduce risks.
Data Collection
Data collection can be performed by deploying
several tools such as API, crawling, and so on. For
texts, it will be gathered based on particular keywords
from chosen individuals who are interested
in the cryptocurrency sector, such as Elon Mask
and Vitalik Buterin. Cryptocurrency, virtual currency,
digital currency, bitcoin, ethereum, and litecoin
are the major search terms. Additionally, in
50
order to capture all behavioral signals from users,
interactions like as likes, shares, retweets, andhashtags
canbe included to the data collecting process.
Data Cleaning
After collecting data from microblogging
sites, it needs to be cleaned. It may include noisy
data leads to inaccurate insights, which will
affect badly on the decisions of cryptocurrencies
users. Dirty data may be classified as misinformation,
missing data, irrelevant data (eg,
retweets, hashtags, URLs), and so on. There are
many proposed techniques to handle these
issues such as machine learning techniques and
missing data imputation techniques.
Data Storing
Hence, the dataset can be stored by different
storage systems including Hadoop and Apache
Spark. For Hadoop, the data is stored by breaking
down the files into large blocks and assorting
them across nodes. Furthermore, Hadoop processes
data in parallel by including code in
nodes to process. Apache Spark is considered
another powerful tool for storing high-dimensional
data sets and has the ability to analyze
data faster than Hadoop.
Categorization
In this step, we categorize unstructured data
into homogenous data by using different techniques
like a fuzzy logic-based approach, a Naı¨ve
Bayes-based approach, term frequency-inverse
document frequency.
Feature Extraction
After categorizing data, we will extract features
for each labeling group of the same topic and combine
all extracted features in one set to be ready for
the feature selection process. Feature extraction is
based on identifying all characteristics of the input
data that maybe valuable for handling thepotential
cryptocurrencies risks. It can be classified into two
types: Structural features and behavioral features.
Structural features include all the words after preprocessing
the post/tweet or comment and behavioral
features that tell us about the mood status
and behavior of the user such as connections number,
sentiment score, and number of hashtags features.
Behavioral features can be identified
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - January/February 2024

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2024
