Discover Internet of Things

Brief Communication

Social media analytics of the Internet of Things

Jim A. Scheibmeir1 · Yashwant K. Malaiya2

Received: 22 June 2021 / Accepted: 13 July 2021

© The Author(s) 2021  OPEN

Abstract
The Internet of Things technology offers convenience and innovation in areas such as smart homes and smart cities. Inter-
net of Things solutions require careful management of devices and the risk mitigation of potential vulnerabilities within
cyber-physical systems. The Internet of Things concept, its implementations, and applications are frequently discussed
on social media platforms. This research illuminates the public view of the Internet of Things through a content-based
and network analysis of contemporary conversations occurring on the Twitter platform. Tweets can be analyzed with
machine learning methods to converge the volume and variety of conversations into predictive and descriptive models.
We have reviewed 684,503 tweets collected in a 2-week period. Using supervised and unsupervised machine learning
methods, we have identified trends within the realm of IoT and their interconnecting relationships between the most
mentioned industries. We have identified characteristics of language sentiment which can help to predict the popularity
of IoT conversation topics. We found the healthcare industry as the leading use case industry for IoT implementations. This
is not surprising as the current COVID-19 pandemic is driving significant social media discussions. There was an alarm-
ing dearth of conversations towards cybersecurity. Recent breaches and ransomware events denote that organizations
should spend more time communicating about risks and mitigations. Only 12% of the tweets relating to the Internet of
Things contained any mention of topics such as encryption, vulnerabilities, or risk, among other cybersecurity-related
terms. We propose an IoT Cybersecurity Communication Scorecard to help organizations benchmark the density and
sentiment of their corporate communications regarding security against their specific industry.

Keywords  Internet of Things · Social media · Cybersecurity · Machine learning · Sentiment analysis · Popularity
prediction

Abbreviations
API Application programming interface
CSV Comma-separated values
DoS Denial of service
IDT Intrusion detection systems
IoT Internet of Things
NRC National Research Council Canada
NLP Natural language processing
QoE Quality of experience
WSS Within-cluster sum of squares

*  Jim A. Scheibmeir, [email protected] | 1Systems Engineering Department, Colorado State University, 400 Isotope Dr,
Fort Collins, CO 80523, USA. 2Computer Science Department, Colorado State University, 1873 Campus Delivery, Fort Collins, CO 80523,
USA.

Discover Internet of Things (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5

13
Vol.:(0123456789)
Brief Communication Discover Internet ofThings (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5

1 Introduction

The Internet of Things (IoT) is an appealing technology that has eased the management of homes through smart appliances
and has enticed industries such as automotive, transportation, and agriculture [1]. IoT was first introduced in 1999 as a tech-
nology concept for solving opportunities within logistics [2]. The IoT phenomenon brings compute from the cloud closer to
people and things [3]. Today, consumers of data are also producers of data. Twitter users tweet nearly 277,000 times every
single minute [4]. The action of liking or retweeting a tweet is yet another data point.
We have collected 684,503 tweets within a two-week period from May 1st, 2021, through May 14th, 2021. Twitter data
has been utilized in several recent research investigations [5–7]. Social media platforms have been found to support access
to information, discuss and solve engineering problems, identify new trends and communicate science to a public audience
[8–12]. We extend the collected Twitter data with metadata using hierarchical clustering techniques and content-based
analysis. The clustering algorithm is informed of proper cluster distribution by the within-cluster sum of squares (WSS) and
average silhouette methods. A content-based analysis is then performed to identify the number of industries, trends, and
technology vendors having a presence in the tweets. Sentiment analysis is carried out for tweets classified towards the indus-
tries and vendor technologies. Factors such as the trend labels, industry labels, and sentiment scores are then used in naïve
Bayes prediction models. We illustrate the relationships, or lack of, between the trends, industries, and technology providers
utilizing network graphs. Section two contains a brief background on the topics of IoT and social media. The research and
analysis methodology are described in detail within section three. Finally, the fourth and fifth sections offer discussion and
conclusion to the research. The main contributions of this research work include:

• Using advanced statistical and machine learning (ML) methods including naïve Bayes, hierarchical clustering, and
natural language processing with sentiment analysis, we evaluate 684,503 contemporary tweets on the topic of the
Internet of Things to shed light on public opinion, technology trends, popular industry usage and the popularity and
sentiment of technology providers in this space.
• We uncover the substantial problem of a lack of cybersecurity discussion within the IoT tweets. No cybersecurity con-
cepts were identified in the top ten trends. Organizations must increase their cybersecurity communication cadence
to meet the risks.
• We analyzed tweets to identify industries where IoT concepts and technology are being discussed. We found health-
care to be the leading industry of mention.
• We propose a new IoT Cybersecurity Communications Scorecard. The scorecard uses a combined index of mention
density and sentiment analysis to provide a benchmark of cybersecurity communication posture scores by industry.
• The top three trends identified within the IoT tweets were data science, machine learning and big data. We performed
a network analysis to identify relationships between trends and industries, such as what industries have the greatest
or least inclusion of trending concepts and technology.
• We evaluate commercial vendors by the sentiment of messages where they are discussed, as well as the volume of
mentions. We provide a positional rank of a selection of IoT commercial technologies based upon this analysis.

We believe our research benefits cybersecurity experts, IoT practitioners, and commercial firms. Cybersecurity practi-
tioners and organizational leaders can utilize our findings and scorecard to benchmark areas of their internal behavior.
Practitioners, such as developers and engineers of IoT systems, can utilized this research to identify trends within the
realm of IoT. Marketing departments of commercial firms benefit from the sentiment analysis and predictive models that
shed light on Twitter user behavior regarding the communication of IoT systems. Our contributions are further discussed.

2 Background work

2.1 Use of social media in research

The Twitter data has been utilized in several recent research investigations [5–9]. The public availability of the tweets
allows researchers to extract valuable conclusions from them [13]. It has been found that the geotagging of twitter
users’ tweets can complement surveys as well as enhance a sampling profile [14]. The same study found that their
survey showed bias towards elderly participants while the Twitter data was biased towards a younger population. The

13
Vol:.(1234567890)
Discover Internet of Things (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5 Brief Communication

researchers utilized these conflicting biases to balance their findings. A study of 640 university students found that the
leading factor for using social media was to search for and access information [8]. An earlier study by Bougie et al. [9] fol-
lowed software engineering groups on Twitter to determine how they utilized the platform. This study found that 23% of
the groups’ tweets were towards software engineering topics. Of that 23% of their total tweets which regarded software
engineering, 62% were towards solving software engineering problems. Another study sought to answer if software
engineering practitioners use and cite scientific research in their blogs; they do not [10]. Rather, software engineering
practitioners utilize social media to become up to date on technology trends [11]. Another research article states that
microblogging serves by linking to web resources, connecting users, and directing users’ attention, as well as offering
another channel for the public communication of science [14].

2.2 Related works on the Internet of Things

Implementing an IoT system requires storage, networks, load-balancing, and analysis tools. According to Atalay and
Angin [15], an IoT solution should utilize network partitions in private clouds which provide partitioning for enhancing
security. Such network partitions could encapsulate the concerns of actuators and sensors, the model of system states,
and the business and program logic. An encryption key management system would be utilized to support encryption
across network enclaves and an intrusion detection system (IDS) could be implemented to identify malicious activity.
The growing interest in IoT and the implementation of the systems have resulted in large cyber-attack surfaces [15].
A few well-known cyber-physical system attacks include the Stuxnet effect on an Iranian uranium enrichment plant [16]
and more recently, a ransomware attack upon Colonial Pipeline that resulted in the gas pipeline being shut down for six
days and a near $5 million payout to the hackers [17]. Another recent example of the threat to cyber-physical systems is
the Solarwinds attack and the resulting 25% of North American electric utilities that were vulnerable [18]. There are also
known consumer exploits including hacked Smart TVs listening to conversations, personal information being extracted
from coffee machines, and security cameras leaking images [19]. Today, IoT implementations may be utilized to carry
medicine, medical samples, and to assist with the management of pandemics [20], furthering the need for rigor and
security in the implementations. In a 2019 research survey of 220 security leaders in industrial and manufacturing, 79%
of respondents indicated they had experienced an IoT cyberattack within that past year [21]. The security aspects of IoT
have the attention of legitimate organizations who seek to enhance the defense as well as the hackers.
The cybersecurity concerns of IoT systems are growing in complexity and have insufficient security solutions [15].
The evaluation of cyber-physical system component vulnerabilities is a challenging task due to the sheer number of
devices and their varied configurations. Common threats include denial of service (DoS) attacks while a common weak-
ness is insecure wireless networking [15]. The complexity of IoT systems and their emergent behavior also complicate
the testing of the systems [22].
To manage the complexity while achieving value and providing security of the system’s assets and users, five best
practices have been suggested by Shi et al. [3]. Good service management of edge computing and IoT systems include
these five considerations:

• Differentiation in device identification to discern specific state metrics such as the health of the specific device
instances.
• Extensibility in the system to allow for replacement endpoints being easily swapped in and out.
• Isolation of access and data via roles and other controlling factors.
• Security/privacy which preserves availability and confidentiality characteristics.
• Optimization of the system and components’ attributes such as cost, latency, or bandwidth.

While Shi et al. [3] mention the optimization of system components including latency and bandwidth, Fizza et al. [23]
dive deeper into optimization stating that existing definitions of quality of experience (QoE) must be renewed with the
autonomous IoT systems in mind. The same research found that if QoE is not considered in autonomous IoT applications,
poor quality of decisions and resulting actions may occur. Motta et al. [24] have examined the IoT-related literature to
find twenty-nine definitions of the concept. Connectivity, a component of QoE, is among the common concepts within
the definitions Motta et al. distilled. From those twenty-nine definitions, they have identified seven key facets. These
facets must be considered when engineering an IoT software system. They include:

13
Vol.:(0123456789)
Brief Communication Discover Internet ofThings (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5

• Connectivity includes the medium for things to connect to implement the IoT paradigm. Connectivity may be chal-
lenged by security concerns or the quality of service.
• Things include the number of heterogeneous tags, sensors, actuators, among other things. There exist challenges of
maintaining the identities of these devices as well as managing their behavior.
• The behavior of IoT systems may include emergent behavior, which is the nonobvious side effects resulting from
the composition of individual parts into a system. The main cause of emergent behavior is due to the complexity of
systems and the human interaction within them [25].
• The smartness of the things within the IoT system relates to how devices are managed, orchestrated, and their allow-
ance and use of autonomous behavior.
• Problem Domain may refer to the industry or specific problem that the IoT software system is built to alleviate.
• Interactivity is not limited to the interaction between things and humans, but also the interaction amongst things
within the IoT system. This implies the importance of interoperability.
• The environment is the context in which an IoT system operates and can also be specific to the problem domain or
implementation.

3 Methodology

3.1 Data acquisition and preprocessing

We wrote and utilized an R program to manage the downloading of tweets from Twitter’s application programming
interface (API). Another R program was created to label the tweets and to perform the content-based analysis. The analysis
begins with preprocessing the tweets including the removal of stop words and usage of word stemming and lemmati-
zation. The analysis includes an identification of trends within IoT discussions. The tweets are labeled for the factors of
popularity (tweets that were liked or retweeted), industry mention, commercial vendor technology mention, and trend
identification. There is an evaluation of sentiment within the labeled tweets. We also analyze the relationships between
the factors of industry and trending terms. A naïve Bayes model is created to determine whether our labeled factors
can predict the content or popularity of the tweets. Using the factors of favorite, industry type, retweet, and IoT vendor
name, we could predict the trend a tweet was referencing with an accuracy of 63.9%. Figure 1 presents our methodol-
ogy in seven steps. The seven steps are carried out in two R programs. The R programs and a compressed CSV file of the
684,503 tweets are available for use and evaluation on a publicly available Gitlab site [26] (Fig. 1).
To perform the collection of tweets, we first created a programming account on the Twitter platform. This account
creation offered the authentication and authorization needed to access the Twitter platform via API. For the first fourteen
days of May 2021, we searched for tweets containing #iot and stored up to 50,000 per day. The search limit was required
as our AWS EC2 server instance is limited to four cores and 32 GB of memory. The impact of the limited server resources
will be described later in this section. By the last day of tweet collection, we had successfully captured 684,503 tweets
containing #iot.

3.2 Number of cluster determination

After data collection, we created a document-term object matrix. The individual words from each tweet were then
cast into the matrix and their frequency of appearance recorded. To determine an ideal number of clusters, we utilized
within-cluster sum of squares (WSS) and the average silhouette methods. However, due to the size of the term matrix
as input into these methods and the restrictions of our compute environment, only samples of the entire tweet corpus
were used to generate the term matrix.
The WSS method will iterate through many generations of k-means clusters. During each iteration through k number
of clusters, the squared distance between a cluster’s observations (within cluster) and the clusters’ centroid are summed
and plotted for the given number of clusters. This is done for all clusters and compared for Euclidean distance over the
iterations. The ideal number of clusters is frequently determined visually, known as the “elbow method” and identified
when the WSS is decreasing and the next increment in cluster generation does not offer much benefit. This is often
visually detected by looking for the “elbow” or the “knee” in the line chart where the WSS has dropped and then flattens.
Figure 2 identifies the knee at four clusters for our dataset of #iot tweets collected over 2 weeks.

13
Vol:.(1234567890)
Discover Internet of Things (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5 Brief Communication

Fig. 1  Seven steps make up

the methodology starting
with an iterative collection of
tweets through labeling and
analysis until the visualization
of the data

Fig. 2  Output of the within-

cluster sum of squares
method to determine the
proper number of clusters to
be generated. The method
indicated that four clusters
were appropriate for the IoT
tweets dataset

13
Vol.:(0123456789)
Brief Communication Discover Internet ofThings (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5

The silhouette method is like the WSS method in that it also generates many iterations of clusters and evaluates them
for a proper k size. The average silhouette evaluation is performed by comparing the silhouette width of each cluster
within an iteration to cluster widths of succeeding iterations having incrementing numbers of clusters. Overall, when
many clusters are found within a small dimension, the width of the clusters (silhouettes) are smaller than if one cluster
was occupying the same space. Thus, when having many small clusters in a dimension that could be optimized by hav-
ing fewer clusters, the average silhouette method will indicate a small average cluster width and an improper number
of k clusters.
Additionally, if clusters are generated as tightly grouped neighbors, then one observation in one cluster will be very
close in distance to an observation in a neighboring cluster. The closeness of observations belonging to different clusters
can indicate that the model suffers too many clusters. A quality number of clusters to generate would be the number
of clusters that optimizes the largest average silhouette width. Ribeiro et al. [27] utilized maximum silhouette scores
in their graph-clustering algorithm to identify groups of terms and their semantics. Their method, and the inclusion
of silhouette scoring, outperformed previous methods. In our research, the silhouette method suggested the proper
number of clusters for our dataset of IoT tweets to be five (as shown in Fig. 3), whereas the WSS method suggested the
proper number of clusters to be four. To ease the execution of algorithms, we utilized R packages factoextra and NbClust.
Unsupervised hierarchical clustering was performed for both four and five cluster outcomes. An agglomerative method
was used. With agglomerative clustering, each observation initializes as its cluster and through iterations is joined with
nodes being the shortest distance away [28]. The difference of trend identification between the different cluster gen-
erations, whether four or five clusters, was not found to be interesting. This is further illustrated within Fig. 4 where the
largest clusters of tweets were cast into word clouds. It is seen that the leading terms are still quite similar despite the
differing number of clusters generated. What was most concerning, whether four or five clusters were generated, was the
lack of any cybersecurity topic as a trending top ten topic. Only 12% of the 684,503 tweets contained any term related
to vulnerabilities, hacking, malware, and other cybersecurity-related terms.

Fig. 3  Output of the silhou-

ette method to determine the
proper number of clusters to
be generated. The method
indicated five clusters as
appropriate for the IoT tweets
dataset

13
Vol:.(1234567890)
Discover Internet of Things (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5 Brief Communication

The tweets were labeled for having inclusion to industry, trend, and commercial vendor technologies. To determine
industry names and search terms, we utilized a list by the International Labor Organization [29]. The tweets were also
evaluated for their sentiment by utilizing the NRC lexicon [30]. Our analysis will be further discussed in the following
section.

4 Findings and discussion

4.1 Unsupervised hierarchical clustering and top trends

Because the WSS and average silhouette methods identified the proper number of clusters for our dataset as four and
five respectively, we generated clusters of tweets for both findings. However, the leading trends identified did not vary
between four and five clusters as illustrated in the word clouds below. Word clouds are basic and intuitive tools that allow
us to evaluate text results for insight [31].
The word cloud on the left is the largest cluster when only four clusters were generated. The word cloud on the right
is the largest cluster when five were generated. We performed a similar trend analysis throughout the cluster creation
and the leading identified trends did not alter. Regardless of the number of clusters created, the top mentioned term
continued to be “data science”. It was closely followed by “machine learning”, and subsequent frequent terms began
dropping off in mention at a greater pace than compared to the first and second most mentioned terms. The mention
analysis of trending topics is illustrated in Fig. 5.

4.2 A small number of cybersecurity mentions within the IoT tweets

Among the trend analysis, in general, what was most concerning was the lack of cybersecurity topics in the list of top
mentioned terms. As illustrated in the following pie chart, only 12% of the 684,503 tweets had any mention of the fol-
lowing stemmed cybersecurity-related terms: cyber, secure, hack, vulnerability, risk, exploit, breach, malware, virus,
ransomware, spyware, worm, trojan, encrypt or phishing (Fig. 6).
When tweets did mention cybersecurity terms, the topics of the three most retweeted conversations included an
industry roundtable discussion [32], a reference to an opinion article about the risk of AI on military technology [33],
and a reference to an article on the risk of AI on national security [34]. Among the most retweeted tweets discussing
cybersecurity, the top three are each a technology being touted to secure IoT implementations.

4.3 An IoT cybersecurity communications scorecard

The absence of frequent cybersecurity discussion within the collection of IoT tweets motivated examining which indus-
tries are communicating about risks the most. To compare the cybersecurity posture of industries based upon the public
discussion found within our collection of tweets, we propose a new IoT Cybersecurity Communication Scorecard. The

Fig. 4  The leading trends

do not include cybersecu-
rity terms nor greatly shift
whether four or five clusters of
our IoT tweets were generated
as indicated by word clouds of
the largest clusters

13
Vol.:(0123456789)
Brief Communication Discover Internet ofThings (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5

Fig. 5  Term frequency is high-

est for data science, followed
closely by machine learning

Fig. 6  Only 12% of the total

collection of IoT tweets had
mention of common cyberse-
curity terms

Balanced Scorecard was introduced by Kaplan and Norton in 1992 and links an organization’s broad performance meas-
urements in four key areas [35]:

• customer perspective
• internal perspective
• innovation and learning perspective
• financial perspective

The purpose in a balanced scorecard is to align the organization to the strategy in areas such as human capital, infor-
mation, and the organizational areas of culture, leadership, and teamwork [36]. A good cybersecurity scorecard helps
improve the information and communication regarding cybersecurity [37]. Organizations have cybersecurity goals to

13
Vol:.(1234567890)
Discover Internet of Things (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5 Brief Communication

be within compliance, protect their business, and to maintain their employees’ and customers’ trust. Cybersecurity is not
just about technology and systems, but also the people and processes that rely on and are a part of the systems [38]. Our
scorecard allows organizations to compare their communication of cybersecurity knowledge, awareness, and training
to a benchmark of public discussion within their industry.
Our IoT Cybersecurity Communication Scorecard assesses posture by comparing the z-scores of density mention and
sentiment scores to the relative averages of all collected tweets. Mention density is the percentage of all IoT tweets that
mention cybersecurity topics. The mention density and sentiment are each normalized by mean and standard deviations
into Z-scores. The z-scores reflect an industry’s position in terms of their cybersecurity mention density and the average
sentiment of all tweets that reference their industry. The z-scores are found by first determining the average percentage
of cybersecurity conversations among all tweets and the average sentiment of all tweets. The standard deviations are
also recorded. The z-scores identify the positive and negative distance to the population’s mean. The posture score is a
combined index of the two z-scores. We gave equal weight in the overall posture score calculation. If an organization
placed significant importance on either the volume or the sentiment of the messages, they could apply custom weights.
Organizations should utilize the scorecard as a benchmark to compare their cybersecurity communication volume and
sentiment to their industry’s scores. For an organization to utilize this scorecard as a benchmark, they must determine
their mention density by dividing the number of corporate cybersecurity communications by the total number of cor-
porate communications and compare to their industry within Table 1. A similar comparison can be done to understand
the positivity and sentiment of their corporate cybersecurity communications.
The leading industry by posture score within this social media analysis was found to be mechanical. Tweets within
the food industry scored the lowest posture. The food sector experiences pressures such as climate change, food price
volatility and food security [29]. We must add cybersecurity risk to this list. Recently JBS USA Holdings, a food manufac-
turer which supplies the United States with roughly one-fifth of their meat supplies, experienced a public, expensive, and
business impacting ransomware attack [39]. Due to the ransomware attack, JBS USA Holdings temporarily shut down
operations in nine beef processing plants and eventually paid a ransom of $11 million [40]. Table 1 provides the density
of cybersecurity messages and their sentiment by industry. The table is sorted by posture rank. The scorecard research
is limited by only comparing the top ten industries by volume.

4.4 Content‑based analysis of industries within the IoT tweets

What is further concerning by the dearth of cybersecurity-related discussions within the collection of IoT-related tweets
is that the top mentioned industry was healthcare. Previous research identified healthcare as one of the lesser influential
industries mentioned in research papers on IoT [41]. Our research and this paper are one effort in shifting that claim. The
top ten mentioned industries are depicted in Fig. 7. It is not surprising to see healthcare leading the mentions as many
countries are still experiencing the COVID-19 pandemic. While collecting these tweets based upon the inclusion of #iot,
4% of the tweets referenced COVID-19. Recent research has discussed the relationship between digital twins, IoT, and
contact tracing technology [42], which could be utilized to help understand the behavior of a pandemic. After healthcare,
the second most mentioned industry within the IoT tweets is commerce followed by financial.

Table 1  Industry Industry Mention density Density Z-score Sentiment score Sentiment Z-score Posture score
cybersecurity scorecard
by mention density and Mechanical 40.7% 2.187 0.041 − 0.033 2.154
sentiment analysis
Automotive 31.1% 1.384 0.024 − 0.194 1.19
Commerce 12.1% − 0.198 0.119 0.699 0.501
Public 12.6% − 0.157 0.077 0.31 0.153
Health 10.3% − 0.353 0.093 0.461 0.108
Financial 5.4% − 0.76 0.128 0.79 0.03
Media 2.3% − 1.018 0.155 1.044 0.026
Transportation 7.6% − 0.519 0.057 0.116 − 0.403
Agriculture 12.9% − 0.13 − 0.038 − 0.783 − 0.913
Food 9.9% − 0.38 − 0.21 − 2.412 − 2.792

13
Vol.:(0123456789)
Brief Communication Discover Internet ofThings (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5

4.5 Network analysis and relationship identification

A network analysis was also performed on the relationships between trends and industries. Fundamental parameters
of a network are its number of nodes, otherwise known as the network size, and the number of edges [43]. We are sur-
rounded by naturally connected structures and networks [44]. Industries and technology trends are no different, as we
confirm with this analysis regarding the health industry connections to all of the top identified IoT trends.
To construct the network graph in Fig. 8, the tweets’ metadata labels were cast as nodes into two tables. The first
table listed every industry and the trend terms (nodes) along with a unique identifier. The second table was a large list
of the industry nodes, a corresponding trend node, and a weight column that indicated the frequency when a tweet
was identified as matching both labels. Utilizing the network and igraph libraries in R, we plotted the node and edge
relationships as the data visualization in Fig. 8. This figure is a network graph that has the most mentioned industry,
healthcare, highlighted as a green network node. Then, red lines which indicate relationships, are drawn to each of the
yellow trending terms given both labels co-exist in single tweet metadata that we created during our preprocessing. As
the image indicates, all trend terms are found in the network of healthcare tweets. As Fig. 4 indicated, serverless was the
least mentioned trending term, yet it too has an inner-tweet relationship to those tweets having reference to healthcare.

4.6 Sentiment analysis of commercial technology providers within the IoT tweets

There are many technology providers which have solutions, offer services, or offer platforms to solve IoT opportunities.
We performed a content-based analysis of technology vendors within the IoT space. To determine the list of IoT vendors
to analyze, we utilized two 2020 research reports by Gartner [45, 46]. We utilized the sentimentr library to determine the
sentiment scores of industry technology providers.
We plotted the technology provider names into a chart having four sections. The four sections of the chart have an x
and y-axis, where the x-axis is the z-score of the tweet sentiments when the vendor is mentioned. The z-score is found
by first determining the sentiment of all tweets that mention the commercial technologies, then calculating the average,
and the standard deviation. Then, the z-score for a given technology vendor is calculated by dividing the commercial
vendor’s mentioned tweet sentiment by the number of standard deviations away from the population’s average senti-
ment. The y-axis is measuring the number of times an IoT technology provider is mentioned in our corpus of tweets.
In general, if a vendor is placed on the upper right area of the chart, that implies that they are widely mentioned and
the sentiment of the tweets that they are mentioned within is above average sentiment. If a vendor is found on the bot-
tom left side of the chart, they would be both lower in popularity and lower in sentiment positivity within this collection of
tweets. Any vendors having less than ten mentions within the tweets were removed from the plot. The dashed blue lines
represent the average mentions and average sentiment scores. The average sentiment of all tweets mentioning these IoT
solution vendors is slightly positive. Use caution when reviewing the chart as the y-axis is intentionally logarithmic. The
logarithmic axis allows the data to pull slightly apart, as though zooming in, for the vendors who have lesser mentions.
The vendor placement can be viewed in Fig. 9.
Amazon’s AWS has the most mentions and the most positive sentiment among the vendors being mentioned within
the IoT tweets. The AWS IoT Core can connect IoT devices to AWS cloud services and AWS offers an IoT SDK for devel-
opment in languages such as Java, JavaScript, or Python. JavaScript was identified as one of the top ten trends in our
analysis. AWS IoT Core product supports message brokering for these protocols [47]:

• Message Queuing and Telemetry Transport (MQTT)

• MQTT over Websockets Secure (WSS)
• Hypertext Transfer Protocol -Secure (HTTPS)
• Long Range Wide Area Network (LoRaWan)

Davra is within the bottom left area of the plot. They have fewer mentions in the analysis and the tweets that do men-
tion them tend to have a lower sentiment than average across all of the analyzed technology vendors. Davra offers an IoT
Platform that has features such as access control to both devices and services, service management features including
edge, cloud, Kubernetes, or container deployments, as well as supporting many different IoT device protocols and data
storage capabilities [48].

13
Vol:.(1234567890)
Discover Internet of Things (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5 Brief Communication

Fig. 7  The top ten mentioned

industry within the collection
of IoT tweets was healthcare
followed by commerce and
then financial

4.7 Predictive modeling based upon our IoT tweet metadata factors

Naïve Bayes has been utilized to accurately forecast crime activities including arson, burglary, and theft [49]. Biology
researchers have successfully applied naïve Bayes modeling to determine the presence of links in protein interaction
networks, although anomaly detection was utilized to increase the accuracy [50]. In our research, we utilize naïve Bayes
models to understand relationships between the IoT trends, the sentiment of the content, industries, and IoT technol-
ogy providers.
Using a naïve Bayes model with a dependent factor of trend type and an independent variable of sentiment, we found
that given a tweet is labeled as towards the trending topic data science, there is a 66.7% probability that the sentiment
of the tweet is positive. Tweets that were labeled as towards the IoT trend of natural language processing (NLP) scored
the second highest in positive sentiment probability at 57.1%. Table 2 notates the conditional probabilities as found by
the model.
A second naïve Bayes model was created to help with understanding which factors influence the prediction of tweets
being retweeted. The industry and trend factors had little impact on a tweet being retweeted. However, the sentiment
did affect the probability of a tweet being retweeted. Given the tweets conveyed either fear or joy would improve the
probability of retweet to 13.0% and 12.4% respectively. A third naïve Bayes model was used to predict which trending
term an IoT tweet may be about. Using the factors of favorite, industry type, retweet, and IoT vendor name, we could
predict the trend a tweet was referencing with an accuracy of 63.9%.

5 Conclusion

There are new microblogs on the topic of the Internet of Things each day. From May 1st, 2021, to May 14th, 2021, we
collected 684,503 tweets by searching Twitter’s API for #iot. While previous research has indicated that healthcare is not
a top-three industry influence on the IoT [41], our research determined healthcare the most widely discussed industry
in public IoT conversations on the Twitter platform. While the healthcare industry requires secured information systems,
only 12% of the tweets within this IoT network analysis referenced cybersecurity concepts. Even less, only 10.3% of the
healthcare related tweets referenced cybersecurity concepts.

13
Vol.:(0123456789)
Brief Communication Discover Internet ofThings (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5

Fig. 8  This network graph

utilizes red arrows that depict
relationships between tweets
towards the healthcare indus-
try, highlighted in green near
the top of the image, and all
of the trending terms which
are lighted in yellow

Fig. 9  AWS has the most mentions and the highest sentiment among our corpus of IoT tweets while the technology company Davra would
have a lesser number of mention and a sentiment less than average

13
Vol:.(1234567890)
Discover Internet of Things (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5 Brief Communication

Table 2  Trending IoT tweet topics having the highest probability of positive sentiment are highlighted in this conditional probability table
Trends (below) Anger Anticipation Disgust Fear Joy Negative Positive Sadness Surprise Trust

AI 0.000 0.214 0.107 0.071 0.000 0.107 0.357 0.036 0.000 0.107
BigData 0.149 0.064 0.000 0.064 0.128 0.106 0.234 0.000 0.064 0.191
DataScience 0.000 0.222 0.000 0.000 0.000 0.000 0.667 0.000 0.111 0.000
DeepLearning 0.100 0.100 0.000 0.000 0.100 0.200 0.300 0.000 0.100 0.100
JavaScript 0.000 0.222 0.000 0.000 0.111 0.000 0.444 0.000 0.000 0.222
MachineLearning 0.045 0.136 0.000 0.091 0.000 0.136 0.455 0.000 0.000 0.136
NLP 0.000 0.143 0.143 0.000 0.000 0.143 0.571 0.000 0.000 0.000
Python 0.000 0.000 0.000 0.500 0.000 0.500 0.000 0.000 0.000 0.000

From this collection of tweets, the most common trend term was data science. A network analysis graph depicted
that every trending term was mentioned within healthcare-related tweets. Whereas for the tweets regarding the
shipping industry, only the trends of AI, big data, and machine learning were related. IoT practitioners should utilize
the network analysis to see how similar organizations are communicating and including technical concepts in their
implementation.
No cybersecurity-related terms or concepts, such as encryption, ransomware, zero-trust, or vulnerabilities, were identi-
fied as trending terms. In general, there was an alarming dearth of conversations towards cybersecurity as only 12% of
the IoT tweets contained any mention of cybersecurity related topics.
The trending terms having the highest probability of positive sentiment in a referencing tweet were data science fol-
lowed by natural language processing. We could predict what trend a tweet was referencing with a 63.9% accuracy. To
reach that level of accuracy in the model we utilized the factors of whether the tweet had been retweeted, marked as a
favorite, and by knowing the industry and vendors being mentioned in the tweet’s text. IoT practitioners need to review
our identified trends for how these technologies can benefit their implementations and end-users. Future research should
include a comparison of the trends we have identified and how they may change over time.
A new IoT Cybersecurity Communication Scorecard was proposed. The posture was scored by the density of
cybersecurity conversations and their sentiment. The top ten mentioned industries were ranked by their posture
using our IoT cybersecurity communication scorecard. The mechanical industry had the highest rated posture. The
scorecard is limited in that in only ranks based on communication regarding cybersecurity and future research is
required to tie the posture score into the many additional areas of securing systems. IT security leaders should uti-
lize this scorecard to benchmark their cybersecurity communication density and sentiment compared to the public
discussions referring to their industry.
Amazon AWS that had the highest average sentiment among the vendors that were considered in this research. It
was also Amazon AWS that was most frequently mentioned in this collection of tweets. Commercial firms can utilize
our research and Fig. 9 to assess competing organizations and improve their social media presence and marketing
messages.
A limitation of this research is that only one microblogging site, Twitter, was utilized for data collection. Another
limitation was the available computing power of our systems. Our experience is that 32 GB of memory is not sufficient
when analyzing 684,503 tweets and thus forces the use of samples within the collection. Specifically, we turned to use
samples when carrying out the unsupervised hierarchical clustering and the naive Bayes models within our methodol-
ogy. There is a need to study time dependent trends that will examine if the communication regarding cybersecurity is
increasing towards acceptable values. Such research will require periodic data collection for a period spanning several
months or a few years.

Acknowledgements  We thank the National Research Council Canada (NRC), Saif Mohammad, and Peter Turney for the creation and mainte-
nance of their sentiment lexicon.

Authors’ contributions  JS conceived the study. JS and YM reviewed the literature. JS collected the data and performed the text analysis. YM
supervised the project. All authors contributed to the writing of the manuscript, read, and approved the final manuscript.

Funding  No funding was received for this research effort.

13
Vol.:(0123456789)
Brief Communication Discover Internet ofThings (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5

Data availability  The datasets analyzed during the study and the R code are available in the GitLab repository, https://​gitlab.​com/​jimsc​heibm​
eir/​socia​lmedi​aanal​ytics​ofiot.

Declarations 

Competing interests  The authors declare that they have no competing interests.

Open Access  This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adapta-
tion, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source,
provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article
are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in
the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will
need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://c​ reati​ vecom
​ mons.o
​ rg/l​ icens​ es/b
​ y/4.0
​ /.

References
1. Girma A. Analysis of security vulnerability and analytics of Internet of Things (IOT) platform. In: Latifi S, editor. Information technology—
new generations. Advances in intelligent systems and computing, vol. 738. Cham: Springer; 2018.
2. Ashton K. That Internet of Things thing. RFiD J. 2009;22:97–114.
3. Shi W, Cao J, Zhang Q, Li Y, Xu L. Edge computing: vision and challenges. IEEE JIOT. 2016;3:637–46. https://​doi.​org/​10.​1109/​JIOT.​2016. ​
25791​98.
4. James J. Data never sleeps 2.0. 2014. https://​www.​domo.​com/​blog/​data-​never-​sleeps-​2-0/.
5. Cruickshank IJ, Carley KM. Characterizing communities of hashtag usage on twitter during the 2020 COVID-19 pandemic by multi-view
clustering. Appl Netw Sci. 2020. https://​doi.​org/​10.​1007/​s41109-​020-​00317-8.
6. Guarino S, Trino N, Celestini A, et al. Characterizing networks of propaganda on twitter: a case study. Appl Netw Sci. 2020. https://​doi.​
org/​10.​1007/​s41109-​020-​00286-y.
7. Tien JH, Eisenberg MC, Cherng ST, et al. Online reactions to the 2017 ‘Unite the right’ rally in Charlottesville: measuring polarization in
Twitter networks using media followership. Appl Netw Sci. 2017. https://​doi.​org/​10.​1007/​s41109-​019-​0223-3.
8. Gómez-García M, Matosas-López L, Ruiz-Palmero J. Social networks use patterns among university youth: the validity and reliability of
an updated measurement instrument. Sustainability. 2020. https://​doi.​org/​10.​3390/​su120​93503.
9. Bougie G, Starke J, Storey M, German DM. Towards understanding twitter use in software engineering: preliminary findings, ongoing
challenges and future questions. In: Web2SE ’11. ACM, New York, NY, USA. 2011;31–36. https://​doi.​org/​10.​1145/​19847​01.​19847​07.
10. Williams A. Do software engineering practitioners cite research on software testing in their online articles? A preliminary survey. In: Pro-
ceedings of the 22nd International Conference on Evaluation and Assessment in Software Engineering 2018 (EASE’18). ACM, New York,
NY, USA, pp 151–156. https://​doi.​org/​10.​1145/​32104​59.​32104​75.
11. Storey M. et al. The impact of social media on software engineering practices and tools. In: Proceedings of the FSE/SDP workshop on
future of software engineering research. ACM. 2010;359–364.
12. Büchi M. Microblogging as an extension of science reporting. Public Underst Sci. 2017;26:953–68. https://​doi.​org/​10.​1177/​09636​62516​
657794.
13. D’heer E, Verdegem P. What social media data mean for audience studies: a multidimensional investigation of Twitter use during a current
affairs TV programme. Inf Commun Soc. 2015;18:221–34. https://​doi.​org/​10.​1080/​13691​18X.​2014.​952318.
14. Martín Y, Cutter S, Li Z. Bridging Twitter and survey data for evacuation assessment of Hurricane Matthew and Hurricane Irma. Nat Hazard
Rev. 2020. https://​doi.​org/​10.​1061/​(ASCE)​NH.​1527-​6996.​00003​54.
15. Atalay M, Angin P. A digital twins approach to smart grid security testing and standardization. In: IEEE international workshop on metrol-
ogy for industry 4.0 & IoT, Roma, Italy, 2020; 435–440. https://​doi.​org/​10.​1109/​Metro​Ind4.​0IoT4​8571.​2020.​91382​64.
16. Langner R. Stuxnet: dissecting a cyberwarfare weapon. IEEE Secur Privacy. 2011;9:49–51. https://​doi.​org/​10.​1109/​MSP.​2011.​67.
17. Turton W, Riley M, Jacobs J. Colonial pipeline paid hackers nearly $5 million in ransom. https://w ​ ww.b​ loomb​ erg.c​ om/n​ ews/a​ rticl​ es/2
​ 021- ​
05-​13/​colon​ial-​pipel​ine-​paid-​hacke​rs-​nearly-​5-​milli​on-​in-​ransom. Accessed 14 May 2021.
18. Ikeda S. Utility regulator says solarwinds backdoor was downloaded by 1/4 of electric utilities on the North American power grid. In: CPO
Magazine. https://​www.​cpoma​gazine.​com/​cyber-​secur​ity/​utili​ty-​regul​ator-​says-​solar​winds-​backd​oor-​was-​downl​oaded-​by-1-​4-​of-​elect​
ric-​utili​ties-​on-​the-​north-​ameri​can-​power-​grid/. Accessed 9 Jul 2021.
19. Srinivas, R. 10 IoT security incidents that make you feel less secure CISO Magizine. https://​cisom​ag.​eccou​ncil.​org/​10-​iot-​secur​ity-​incid​
ents-​that-​make-​you-​feel-​less-​secure/. Accessed 10 Jan 2020.
20. Chen CW. Drones as internet of video things front-end sensors: challenges and opportunities. Discov Internet Things. 2021. https://​doi.​
org/​10.​1007/​s43926-​021-​00014-7.
21. Hearn M, Rix S. Cybersecurity considerations for digital twin implementations report. Industrial Internet Consortium. Journal of Innova-
tion. 2019. https://​www.​iicon​sorti​um.​org/​news/​joi-​artic​les/​2019-​Novem​ber-​JoI-​Cyber​secur​ity-​Consi​derat​ions-​for-​Digit​al-​Twin-​Imple​
menta​tions.​pdf.
22. Kim H, Ahmad A, Hwang J, Baqa H, Le Gall F, Ortega MAR, Song J. IoT-TaaS: towards a prospective IoT testing framework. IEEE Access.
2018;6:15480–93. https://​doi.​org/​10.​1109/​ACCESS.​2018.​28024​89.
23. Fizza K, Banerjee A, Mitra K, et al. QoE in IoT: a vision, survey and future directions. Discov Internet Things. 2021. https://​doi.​org/​10.​1007/​
s43926-​021-​00006-7.

13
Vol:.(1234567890)
Discover Internet of Things (2021) 1:16 | https://doi.org/10.1007/s43926-021-00016-5 Brief Communication

24. Motta R, de Oliveira K, Travassos G. On challenges in engineering IoT software systems. In: Proceedings of the XXXII Brazilian symposium
on software engineering. ACM, NY, USA. 2018; 42–51. https://​doi.​org/​10.​1145/​32662​37.​32662​63.
25. Grieves M, Vickers J. Digital twin: mitigating unpredictable, undesirable emergent behavior in complex systems. Cham: Springer Inter-
national Publishing; 2017. p. 85–113.
26. Scheibmeir J. Code and data files. Gitlab. 2021. https://​gitlab.​com/​jimsc​heibm​eir/​socia​lmedi​aanal​ytics​ofiot.
27. Ribeiro E, Teixeira A, Ribeiro R, et al. Semantic frame induction through the detection of communities of verbs and their arguments. Appl
Netw Sci. 2020. https://​doi.​org/​10.​1007/​s41109-​020-​00312-z.
28. Hastie T, Tibshirani R, Friedman J. The elements of statistical learning: data mining, inference, and prediction. Berlin: Springer; 2009.
29. International Labor Organization. Industries and sectors. https://w ​ ww.i​ lo.o
​ rg/g
​ lobal/i​ ndust​ ries-a​ nd-s​ ector​ s/l​ ang-- e​ n/i​ ndex.h
​ tm. Accessed
16 May 2021.
30. Mohammad S, Turney P. Emotions evoked by common words and phrases: using mechanical turk to create an emotion lexicon. In: Pro-
ceedings of the NAACL-HLT 2010 workshop on computational approaches to analysis and generation of emotion in text, LA, California.
2010.
31. Altuncu M, Mayer E, Yaliraki S, et al. From free text to clusters of content in health records: an unsupervised graph partitioning approach.
Appl Netw Sci. 2019. https://​doi.​org/​10.​1007/​s41109-​018-​0109-9.
32. [@#100DaysOfCode]. Sponsored roundtable: artificial intelligence tech—SecurityInfoWatch. May 14, 2021. Twitter.
33. [@HubOfML]. Opinion: artificial intelligence’s military risks, potential—GovTech May 14, 2021. Twitter.
34. [@#100DaysOfCode]. Artificial intelligence and the future of national security—ASU Now May 12, 2021. Twitter.
35. Kaplan RS, Norton DP. The balanced scorecard-measures that drive performance. Harv Bus Rev. 1992;70:71–9.
36. Herath T, Herath H, Bremser WG. Balanced scorecard implementation of security strategies: a framework for IT security performance
management. Inf Syst Manag. 2010;27(1):72–81.
37. Whalen DT, Bell G. Cyber security risk: key areas of focus. Direct Boards. 2014;38(3):14.
38. Bowen BM, Devarajan R, Stolfo S. Measuring the human factor of cyber security. In: Proceedings of the 2011 IEEE International Conference
on Technologies for Homeland Security (HST), 2011;230–235. https://​doi.​org/​10.​1109/​THS.​2011.​61078​76
39. Bunge J. 2021. WSJ news exclusive | JBS paid $11 million to resolve ransomware attack. In: The Wall Street Journal. https://​www.​wsj.​com/ ​
artic​les/​jbs-​paid-​11-​milli​on-​to-​resol​ve-​ranso​mware-​attack-​11623​280781. Accessed 6 Jul 2021.
40. Lerman R. 2021. JBS paid $11 million in ransom after hackers shut down meat plants. In: The Washington Post. https://​www.​washi​ngton​
post.​com/​techn​ology/​2021/​06/​09/​jbs-​11-​milli​on-​ransom/. Accessed 7 Jul 2021.
41. Dachyar M, Zagloel T, Saragih LR. Knowledge growth and development: internet of things (IoT) research. Heliyon. 2019;5:2006–18. https://​
doi.​org/​10.​1016/j.​heliy​on.​2019.​e02264.
42. Scheibmeir J, Malaiya YK. Multi-model security and social media analytics of the digital twin. ASTEJ. 2020;5:323–30.
43. Smith KM, Escudero J. Normalised degree variance. Appl Netw Sci. 2020;5:32. https://​doi.​org/​10.​1007/​s41109-​020-​00273-3.
44. Kolaczyk ED, Csárdi G. Statistical analysis of network data with R, vol. 65. New York: Springer; 2014.
45. Goodness E, Velosa A, Friedman T, Berthelsen E, Kim S, Havart-Simkin P, Thielemann K. Magic quadrant for industrial IoT platforms. Gartner
Inc. Oct. 19, 2020. https://​www.​gartn​er.​com/​doc/​39919​52. Accessed 16 May 2021.
46. Wong J, Leow A, Batchu A. Magic quadrant for multiexperience development platforms. Gartner Inc., July 7, 2020. https://​www.​gartn​er.​
com/​doc/​39872​01. Accessed 16 May 2021.
47. AWS. What is AWS IoT? AWS. https://​docs.​aws.​amazon.​com/​iot/​latest/​devel​operg​uide/​what-​is-​aws-​iot.​html. Accessed 1 Jun 2021.
48. Davra. IoT platform: internet of things platform—davra IoT platform. Davra. https://​davra.​com/​iot-​platf​orm/. Accessed 1 Jun 2021.
49. Niu X, Elsisy A, Derzsy N, et al. Dynamics of crime activities in the network of city community areas. Appl Netw Sci. 2019;4:127. https://​
doi.​org/​10.​1007/​s41109-​019-​0239-8.
50. Singh KV, Vig L. Improved prediction of missing protein interactome links via anomaly detection. Appl Netw Sci. 2017;2:2. https://​doi.​
org/​10.​1007/​s41109-​017-​0022-7.

Publisher’s Note  Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

13
Vol.:(0123456789)