Revisiting Social Media Tie Strength in the Era of Data Access

Restrictions

Jayesh Prakash Gupta

1 a

, Hannu K

arkk

ainen

1 b

, Osku Torro

1 c

and Raghava Rao Mukkamala

2,3 d

Unit of Information and Knowledge Management, Tampere University, Tampere, Finland

Department of Digitalization, Copenhagen Business School, Copenhagen, Denmark

Department of Technology, Kristiania University College, Oslo, Norway

Keywords:

Data Access Policy, API, Tie Strength, Social Media, Tie Strength Models, Weak Ties, Strong Ties.

Abstract:

The strength of social ties has an impact on how information is transferred and processed in a social network.

Many studies have used social media data to evaluate tie strength. However, many of these studies were done

at a time when social media data could be accessed legally without using the social media platform API. In

the past few years, there have been signiﬁcant changes in the data access policies of these platforms, which

has led to a considerable reduction in the possibilities of using social media data for tie strength evaluation.

The paper aims to study the impact of the data access policy changes of major social media platforms on the

existing social media based tie strength models. The ﬁndings of this study show that the existing social media

based tie strength models can no longer be utilized in their current form. Our study suggests that there is either

a need to modify the existing social media based tie strength models or to develop new social media based tie

strength models that reﬂect the recent changes in the data access policies.

1 INTRODUCTION

The concept of strong and weak ties was introduced

by Granovetter in his seminal study The Strength of

Weak Ties (Granovetter, 1973). Personal social ties

and networks have been demonstrated as quintessen-

tial in the effective information and knowledge trans-

fer. Strong and weak social ties have been stud-

ied in many contexts, including knowledge and in-

formation management, as well as organization sci-

ence In more detail, weak ties have been found to be

of great importance in encouraging information ex-

change and avoiding redundancy, while strong ties

have been found more likely to facilitate to tacit

knowledge transfer. (Gupta et al., 2016) In addi-

tion, knowledge workers across different work do-

mains have been found to utilize weak ties to improve

their work efﬁciency in different manners such as lo-

cating new useful knowledge (Zhang et al., 2017).

Various models and algorithms to evaluate the tie

strength has been developed during the years to detect

https://orcid.org/0000-0003-4043-4818

https://orcid.org/0000-0003-4753-4416

https://orcid.org/0000-0003-0706-5010

https://orcid.org/0000-0001-9814-3883

weak and strong ties from a variety of data sources.

Over the decades, the concept of tie strength has been

used to study various social phenomena, such as inno-

vation and creativity, knowledge transfer, information

diffusion, and content sharing (Gupta et al., 2016).

At the same time, the tie strength measurement has

been extended from its original use at an interpersonal

level to a group-level, organizational level, and inter-

organizational level, as well (see (Zhang et al., 2017)).

In the recent decade, the maturation of various so-

cial media platforms has provided new avenues for in-

formation and knowledge transfer. New models and

approaches for detecting social ties and for the evalu-

ation of tie strength has been created, which make use

of various types of social media data. In such stud-

ies, social media has been found as very prominent

in tie detection, and tie strength evaluation and many

tie strength evaluation models have been coined for

the tie strength( e.g., Gilbert and Karahalios (2009);

Jones et al. (2013)). During the last few years, there

have been signiﬁcant changes in the data access poli-

cies of major social media platforms, such as Twit-

ter and Facebook (Hogan, 2018). These changes are

due to various reasons, including changes in platform

companies’ business models, changes in data regula-

tions such as the EU-originated GDPR, as well as the

Gupta, J., Kärkkäinen, H., Torro, O. and Mukkamala, R.

Revisiting Social Media Tie Strength in the Era of Data Access Restrictions.

DOI: 10.5220/0008067501870194

In Proceedings of the 11th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2019), pages 187-194

ISBN: 978-989-758-382-7

187

public exposure of unethical conduct related to the use

of social media users’ data. The aforementioned has

undoubtedly impacted the usefulness of social media

data in the evaluation of tie strength. Most of the pre-

vious studies and the predictors related to social me-

dia datamade use of private data, which is no longer

legally accessible from major social media platforms

(Gupta et al., 2016).

Moreover, while many of the most cited social

media-based tie strength models were created before

many signiﬁcant data access changes had taken place,

this has presumably affected the usefulness and ac-

curacy of the models, the models’ tie strength di-

mensions, as well as the usefulness of individual tie

strength predictors. There is no study that explored

the impact of data access changes on the tie strength

evaluation, and this paper aims to shed light on the

impact of data access changes on the tie strength eval-

uation. The following is our research question.

RQ: “How the data access changes of major so-

cial media platforms impacted the utility of existing

social media tie strength evaluation models?”

This study is not only relevant for the overall

purposes of tie strength but also for the develop-

ment of better social recommendation systems based

on strong and weak ties (see Huhtam

aki and Olsson

(2018)). The structure of the paper is as follows. In

sections 2 and 3, we ﬁrst introduce the concept of tie

strength, then social media based tie strength models.

Then in section 4, we explain the methodology of the

paper and section 5 will present our ﬁndings. Finally,

in section 6, we will discuss the conclusions and fu-

ture work.

2 CONCEPT OF TIE STRENGTH

The concept of tie strength was originally introduced

by Mark Granovetter (1973) in his seminal study “The

Strength of Weak Ties”. According to him tie strength

can be deﬁned as ”a (probably linear) combination of

the amount of time, the emotional intensity, the inti-

macy (mutual conﬁding), and the reciprocal services

which characterize the tie” (Granovetter, 1973). In

the original deﬁnition provided by Granovetter, the tie

strength evaluation was used to understand the differ-

ent interpersonal relationships. In other words, the

concept of tie strength provided the degree of close-

ness between two individuals (Gupta et al., 2016).

However, Granovetter left the more precise deﬁnition

of tie strength to future work.

In general, strong ties are people whom you trust

and whose social circles tightly overlap with your so-

cial circle. In the personal context, the strong tie is the

people with whom you have a long relationship his-

tory, interact regularly and share different life expe-

riences, such as family members. In the professional

context, strong ties might be people with whom you

work in a project or in the same group, exchange fre-

quent information about work tasks and ask for per-

sonal advice. Strong ties provide emotional support

and are more stable and easy to rely upon. In the pro-

fessional context, as well, people rely on their strong

ties for protection and comfort in situations of uncer-

tainty. (Granovetter, 1973; Gupta et al., 2016)

On the other hand, weak ties are acquaintances or

unfamiliar individuals. Weak ties have been found to

provide access to novel information and to help in the

diffusion of new ideas, helping to provide also new

knowledge for individuals and organizations (Gilbert

and Karahalios, 2009; Granovetter, 1973).

Over the past decade, the rise and popularity of

social media have given rise to new ways to detect,

establish, and manage ties online. The collective pro-

cess of production, consumption, and diffusion of in-

formation on social media are starting to reveal a sig-

niﬁcant portion of human social life. (Gupta et al.,

2016) Thus, the availability and access to social me-

dia have led to many new research opportunities in

the area of tie strength related research. This has re-

sulted in many studies that use social media data to

identify different kinds of online ties by trying to pre-

dict the tie strength of these online relationships. In

this paper, we refer to tie strength calculated using

social media data as Social Tie Strength. However,

there has been a gradual shift in terms of how these

social media platforms share and provide access to

their data for different purposes, including research.

Hence, there is a need to understand how these data

access policy changes of the social media platforms

can impact the tie strength models.

3 INFLUENCE OF API ON

SOCIAL TIE STRENGTH

3.1 Social Tie Strength Models

In this section, a brief description of the major social

tie strength evaluation/calculation is provided. First,

we will describe the kind of social media platform, the

kind of social media data that was used by the model,

and then how the social media data for these studies

were collected. These social tie strength models were

shortlisted based on the criteria deﬁned in section 4.

and based on that criteria, a total of four different so-

cial tie strength models are described below.

KMIS 2019 - 11th International Conference on Knowledge Management and Information Systems

188

3.1.1 Model by Gilbert and Karahalios

The ﬁrst social tie strength model was developed

by Gilbert and Karahalios (2009). This was the ﬁrst

model that tried to use social media data to predict

tie strength between social media users. According to

the deﬁnition of tie strength provided by Granovetter,

tie strength has different dimensions that can be op-

erationalized using different measures and predictors.

In this paper, these different measures and predictors

of tie strength were operationalized using the differ-

ent functionalities/features that were available on the

social media platform, especially for Facebook. A to-

tal of 74 Facebook variables were used as predictors

to measure the social tie strength. The different di-

mensions of tie strength (written in capital letters) and

some of the predictors (written in small letters) used

in this model are also shown in Fig. 1. This model

used the explicit relationship data of Facebook friends

and other explicit Facebook friendship related data

of experimental study participants to calculate the tie

strength. The social media data for this study was col-

lected by crawling the Facebook page and proﬁle data

of the study participant. In this study, the ﬁnal devel-

oped social tie strength model was a linear combina-

tion of the predictive variables, the pairwise interac-

tion of the predictive variables and the network struc-

ture where the network structure was based on the ex-

plicit Facebook friendship data that was crawled from

Facebook. This model was later transformed and ap-

plied to tie strength calculation using Twitter data.

Many of the Facebook predictors, which were used

in the previous model, were substituted with similar

predictors from Twitter data. However, this Twitter

data based tie strength model did not have predictors

from all the different tie strength dimensions (Gilbert,

2012).

Figure 1: Tie strength dimensions and predictors used in tie

strength model by Gilbert and Karahalios (2009).

3.1.2 Model by Kahanda and Neville

Kahanda and Neville (2009) developed a supervised

learning model for social tie strength calculation with

the goal of detecting strong and weak ties. In practice,

this model was able to perform the binary task of pre-

dicting whether or not a relationship was strong. The

model was based on using Facebook data, which was

divided into four different kinds of graphs: friendship

graph, wall graph, picture graph, and group graph.

From these graphs, a total of 50 different features

were constructed and used for classiﬁcation of the re-

lationship as being strong or not. This model also

used explicit relationship data related to Facebook

friends. A publicly available social media dataset -

Purdue Facebook network was used to carry out this

study. However, this data is no longer available due to

the change in Facebook terms of service.

3.1.3 Model by Xiang et al.

Xiang et al. (2010) developed an unsupervised learn-

ing model for social tie strength to infer relationship

strength based on proﬁle similarity and interaction ac-

tivity. The model was based on the principle of ho-

mophily i.e., people tend to form relationships with

similar kinds of people. Hence, the model assumed

that the stronger the tie, the higher the similarity. The

model tried to model the tie strength as the hidden

cause of user proﬁle similarities and user interactions.

The model used social media data, which was divided

into four kinds of graphs: friendship graph, top-friend

graph, wall graph, and picture graph. These graphs

were used to evaluate the tie strength between the so-

cial media users of this study. This model also used

the explicit relation data, which was Facebook friends

in case of the Facebook dataset and the Linkedin net-

work data in case of the Linkedin dataset. This study

used two different social media datasets. The ﬁrst

dataset was a proprietary dataset from Linkedin.com.

The second dataset was a public dataset from the

Purdue Facebook network. However, this data is no

longer available due to the change in Facebook terms

of service.

3.1.4 Model by Jones et al.

Jones et al. (2013) developed a logistic regression

based-social tie strength model, which could deter-

mine how real-world tie strength can be inferred from

easily measurable online behavior and demographics

using Facebook. The created model was successful

in differentiating between the closest friend from not-

closest friends in real world relationships. Logistic

regression provides the probability based on the pro-

vided data features, which in this case was to give

the probability about whether two Facebook users of

the study were close friends in the real-world. The

model was based on only one feature, which was the

Revisiting Social Media Tie Strength in the Era of Data Access Restrictions

189

sum of all the interactions between the users known

as summed interaction. This model also used some

private Facebook data of the study participant like the

pokes, messages. The social media data for this study

was collected using the Facebook Id of the study par-

ticipants. The study did not specify whether the data

was crawled directly or using graph API.

3.1.5 Recent Social Tie Strength Models

In recent years, more social tie strength models have

been developed. Many of these models are also

listed in the paper by Liberatore and Quijano-Sanchez

(2017). However, other than the above mentioned

four social tie strength models, all the other social

tie strength models are based on the minor modiﬁ-

cation of the earlier existing social tie strength mod-

els. For example, the social tie strength model by

Quijano-S

anchez et al. (2014) and social tie strength

model by Fogues et al. (2018) are based on the minor

modiﬁcation of the original social tie strength model

by Gilbert and Karahalios (2009). Thus based on

the existing literature related to the social tie strength

model, the above mentioned four social tie strength

models are the current state of the art. These four

social tie strength models can be considered as repre-

sentative of all the existing different social tie strength

models.

From the description of the different social tie

strength model, it can be observed that social tie

strength models by Gilbert and Karahalios (2009);

Kahanda and Neville (2009) and Xiang et al. (2010)

used explicit social media relationship data in or-

der to develop their social tie strength model. Also,

the models by Gilbert and Karahalios (2009) and by

Jones et al. (2013) used the private social media data

of the study participants. Both the explicit relation-

ship data and private user data are no longer accessi-

ble from social media. Thus, we want to understand

how the changes in social media data access have

impacted the utility of the current social tie strength

models.

3.2 Impact of APIs on Data Access

The growth of the data on the internet in general and

social media, in particular, resulted in the develop-

ment of certain methods and standardization which

could be used to access the data from these social

media platforms resulting in the creation of applica-

tion program interface commonly referred to as API.

API became the de facto means by which data could

be sent between devices and could be accessed from

these platforms. The second most important aspect

along with the API was the aspect of authentica-

tion, i.e., the authentication that servers use to iden-

tify clients (i.e., third parties that work on behalf

of users) and provide personalized access based on

which client is requesting what data. (Burgess and

Puschmann, 2014; Hogan, 2018) The combination of

API and authentication, resulted in the creation of

a speciﬁc kind of API known as authenticated API.

These authenticated API could now be used as a tech-

nological gatekeeper to the data stored on the social

media platforms (Hogan, 2018).

The initial introduction of authenticated API was

done to handle the data integrity and data access in a

more technically efﬁcient manner by the social media

platforms. However, with the growth and prolifera-

tion of these social media platforms, these platforms

became a signiﬁcant and vast source of user data. The

concept of big data and its commercial value was no

longer a buzzword anymore, and the social media was

now actually one of the primary examples of how big

data could be used for commercial gains. Social me-

dia data became one of the most critical areas of the

rapidly growing data market. The companies that di-

rectly collect and proﬁt from social media data such

as social media platforms like Facebook and Twit-

ter and also third-party sellers of social media data

like Gnip and Datasift attracted massive valuations.

The business models of these companies moved to-

wards providing privileged access to the social me-

dia data, and the resulting valuable insights which

could be gained from this user-generated social media

data. (Burgess and Puschmann, 2014) Researchers,

analysts, and consultants suggested that advanced sta-

tistical techniques could not only be used to analyze

social media data but also make predictions using so-

cial media data for multiple use cases in very different

ﬁelds. Thus, social media platforms could now use

their role as the social media data provider in order to

drive their business models.

The social media data was now the most valuable

asset of these social media platforms and their con-

trol of how and who could access this data was essen-

tial to their business model. The change in the busi-

ness model of social media platforms has resulted in

the change in the policy of the social media platforms

from the initial era of Web 2.0 where the user had the

power to a more media-centric business model rely-

ing ﬁrstly on advertising and corporate partnerships

and, secondly on reselling the data produced collec-

tively by the platform’s millions of users (Burgess

and Bruns, 2012). This shift has been realized practi-

cally in the architecture of the platform using the so-

cial media platform authenticated API and associated

policies, affecting the ability of third-party develop-

KMIS 2019 - 11th International Conference on Knowledge Management and Information Systems

190

Table 1: The major changes in the data access policy of Facebook and its impact on the tie strength related research.

Major changes to

Facebook data ac-

cess policy

When Overall impact to data access Impact on Tie strength calculation

Introduction of

Open Graph API

v1.0

2010

• Systematic way to directly access data

from Facebook.

• Every data item was assigned a unique id.

• Any Facebook app could access all user’s

Facebook friends and their friend of friend.

• Enabled systematic collection of large

scale tie strength related data including

explicit relationship data (Facebook

friends) using the unique id.

Introduction of

Open Graph API

v2.0

2014

• An app could only access user’s friends if

those friends also authorize the app.

• Facebook replaced the like button with

reactions button.

• Tie strength related data like explicit

relationship data and other personal user

data no longer accessible.

Introduction of

Open Graph API

3.0

2018

• Access to data from open Facebook group

and Facebook pages severely restricted.

• No unique id for same user on different

Facebook groups and pages.

• Individual approval for every app giving

more control to Facebook.

• Tie strength calculation using public

data (open Facebook groups and pages)

not possible due to restricted data access.

• Severely limit the collection of tie

strength predictor related data about

user due to lack of unique user id.

ers, users, and researchers to access, exploit or inno-

vate upon the social media data of these social media

platform (Burgess and Puschmann, 2014). Hence, it

is essential to take into account this crucial factor of

data access and its impact on the social media related

research in general and speciﬁcally towards the so-

cial tie strength related research. The current study

looks speciﬁcally into how the changes in the data ac-

cess policy of the social media platforms (speciﬁcally

Twitter and Facebook) have impacted the previous re-

search related to social tie strength.

4 METHODOLOGY

In this section, how this research was carried out as

explained. This research was carried out pragmati-

cally in the following three different phases. In the

ﬁrst phase, the identiﬁcation and tracking of the sig-

niﬁcant data access policy changes were done mainly

by analyzing the API changelogs of Facebook and

Twitter. In the second phase, the major social tie

strength models were identiﬁed using particular se-

lection criteria from the webofscience database. In

the ﬁnal phase, the data access policy changes iden-

tiﬁed in phase one were used to study the impact of

these changes on the utility and applicability of the

major social tie strength models identiﬁed in phase

two. This process is also illustrated in Fig 2.

The ﬁrst phase consisted of identifying and track-

ing the signiﬁcant changes in terms of the data access

policy of social media platforms that have taken place

since the rise of social media platforms. During this

phase, we decided to focus on the two most signif-

icant social media platforms, which were Facebook

and Twitter. Another important reason for selecting

Figure 2: Different phases in conducting this study.

these two social media platforms was the fact that

these are the most studied platforms in social media

research. In order to track and identify the major API

changes of these two social media platforms, the API

changelogs of Facebook and the API changelogs of

Twitter were accessed and analyzed. Along with the

API changelogs of the social media platforms, aca-

demic articles related to the changes in the API in so-

cial media platforms were also searched on the we-

bofscience database, which provided some relevant

results (e.g., Burgess and Puschmann (2014); Burgess

and Bruns (2012)). The API changelogs of the so-

cial media platforms and the relevant academic arti-

cles were used in identifying the major data access

policy-related changes.

In the second phase, the major social tie strength

models were identiﬁed. The tie strength models

were shortlisted based on a particular selection cri-

terion. These criterion were that: the tie strength

model should be based on the original deﬁnition of tie

strength (hence studies like De Meo et al. (2014) were

excluded as this study changed the basic deﬁnition of

tie strength); the tie strength model should have an in-

dependent mathematical formulation; should use only

social media data for tie strength calculation(e.g., tie

strength models by Onnela et al. (2007) and by Mat-

tie et al. (2018) were excluded since they used call

log data and not social media data); should not just be

Revisiting Social Media Tie Strength in the Era of Data Access Restrictions

191

Table 2: The major changes in the data access policy of Twitter and its impact on the tie strength related research.

Major changes to

Twitter data ac-

cess policy

When Overall Impact to data access Impact on tie strength calculation

Introduction of

Restrictive API

structure

2011

• Twitter streaming API only provides a

certain small percentage of overall Twitter

live stream data.

• No free access to all the past tweets and

restriction on amount of data that can be

accessed from Twitter.

• Signiﬁcant restrictions to the volume

and time period for which the user data

could be accessed and collected.

• Streaming API reduced the volume and

general representation of live data which

could be collected for tie strength .

Introduction of

New Terms of

Service

2012

• New rules that favour only large

institutions or corporations to collect data.

• Restricting large scale data collection

(including Twitter metadata & historical

data) by apps through limited access tokens.

• Push for buying Twitter data ofﬁcially.

• Limited access tokens heavily restrict

speed, duration & volume of tie strength

related data collection from REST API.

• Limited or no access to historical

explicit relationship data & other tie

strength related predictors .

Introduction of

Premium API and

New API

restrictions

2018

• New Premium API resulting in reduced

access to free REST API & shutting

down all data re-sellers.

• Data access (REST API) limited to 7 days.

• Individual approval for all apps.

• Access to the longitudinal tie strength

related data was severely impacted as the

time window for collecting the historical

Twitter data was reduced to just 7 days.

a minor modiﬁcation of an older tie strength model

(e.g., the tie strength model by Fogues et al. (2018)

was excluded as it was just a minor modiﬁcation

of Gilbert and Karahalios (2009) model; and other

studies had cited the tie strength model. Based on

these selection criteria, a total of four major social

tie strength models were identiﬁed from the webof-

science database.

In the third phase, the identiﬁed major data access

policy changes of the social media platform of phase

one were used to study the impact of these data access

changes on the major social tie strength models iden-

tiﬁed in phase two. This allowed the third phase to

identify the major impacts the data access policy has

had on the tie strength calculation research based on

social media. This resulted in the creation of table 1

and table 2.

5 FINDINGS

In this section, the results are presented in the form

of two tables (see Table 1 and Table 2). The structure

of Table 1 and Table 2 are similar. The ﬁrst column

of both the tables shows what caused a major change

in the data access policy of the social media platform.

The second column gives the year from when these

changes were introduced. The third column provides

the overall impact of these changes in the data ac-

cess policy had to access data from this social me-

dia platform. The third column explains impacts to

data access related to what kind of data and meta-

data could be accessed or became inaccessible, the

volume of data that could be accessed, the period for

which data could be accessed and also the purpose for

which the data could be accessed from the social me-

dia platform. The ﬁnal column of the tables explains

the speciﬁc impact these data access changes had on

the tie strength calculation using social media data. It

can be seen from Table 1 and Table 2 that the ﬁrst ma-

jor changes related to data access policy to Facebook

and Twitter happened in 2010 and 2011, respectively.

Facebook and Twitter introduced these changes, the

primary method of accessing the data from these so-

cial media platforms was not just limited to using the

API of the respective platforms.

From Table 1, it can be seen that the initial in-

troduction of open graph API v1 made it very easy

to get access to large scale tie strength related data

including the explicit relationship data about Face-

book Friends. This data has been used by all the tie

strength models based on social media data, which

were described in section 3.1. However, this access

to explicit relationship data about Facebook Friends

was practically removed with the introduction of open

graph API v2. The introduction of open graph API v3

has even made an access and the linking of the public

Facebook information about a user almost impossi-

ble. Thus, it can be seen that after the introduction

of these changes, it is not possible to directly use the

tie strength predictors which were used in social tie

strength models that were described in section 3.1.

From Table 2 it can be seen that when Twitter

introduced streaming API and revoked access to the

unlimited historical Twitter data which could be ac-

cessed by researchers, it massively reduced the abil-

ity of the researchers to get historical Twitter data.

There was a further decrease in the volume of data

that could be collected from Twitter with the intro-

duction of a limited number of access tokens. For the

KMIS 2019 - 11th International Conference on Knowledge Management and Information Systems

192

tie strength model based on social media in general

and speciﬁcally the tie strength model developed by

Gilbert (2012), it was no longer possible to get ac-

cess to some signiﬁcant tie strength predictors like

the historical number of followers (explicit relation-

ship data), direct messages and some other relevant

predictors. The access to explicit relationship data

is essential for the model by Gilbert as it relies on it

for accurate tie strength predictions. The introduction

of Premium API by Twitter has further reduced the

amount of longitudinal tie strength related data that

can be accessed.

Thus, it can be seen from Table 1 and Table 2 that

it is not possible to directly use the major social tie

strength models described in section 3.1 as the access

to explicit relationship data which is used by all these

tie strength models is no longer possible.

6 DISCUSSION AND

CONCLUSIONS

Based the analysis of this study regarding the ma-

jor social media platforms’ data access policies and

major-related changes, it seems evident that there

have been a lot of signiﬁcant changes speciﬁcally in

Facebook’s and Twitter’s data access. As seen in the

previous analysis, these changes are signiﬁcantly im-

pacting the social media based tie strength research

in general, as well as the usefulness and accuracy of

currently existing social tie strength models.

In more detail, ﬁrst it can be seen from the pre-

vious analysis that all the current social tie strength

models rely heavily on using either explicit rela-

tionship data (e.g., friendship data) or private user

data (e.g., direct messages) or both from social me-

dia platforms. However, this kind of social me-

dia data is either no longer accessible (e.g., Face-

book Friends, direct messages) or severely restricted

(e.g., follower/followee)from access from social me-

dia platforms.

Secondly, it can be observed that when some of

the initial social tie strength models were created

(e.g., Gilbert and Karahalios (2009)) there existed

many ways to legally collect social tie strength re-

lated data from social media platforms. However,

now, APIs are the only legitimate way of accessing

and collecting the data from social media platforms

(speciﬁcally Facebook and Twitter).

Third, researchers cannot access the historical, so-

cial media data (e.g., past ﬁve years daily Twitter fol-

lower count). The current social tie strength models

were built using a lot of historical longitudinal social

media data. This historical data allowed a more in-

depth and accurate tie strength evaluation and calcu-

lation, which is no longer possible with the current

data access policies of the social media platforms.

To the best of our knowledge, this is the ﬁrst study

that analyses how the changes in the data access poli-

cies of the social media platforms have impacted the

usefulness and application of the existing social tie

strength models.

6.1 Implications to Research

The results of this study have some important implica-

tions for the research and researchers of tie strength.

First, overall, the relatively recent changes in major

social media platforms’ policies mean that in the case

of all the currently existing social tie strength mod-

els, it is impossible to use these models directly, as

such, for accurate tie strength prediction and evalua-

tion. This is because all these social tie strength mod-

els rely on explicit relationship data and private data

from social media, which is no longer accessible.

Second, we are aware that many factors like reg-

ulations (such as GDPR), platform business model

changes and experiments, public exposure of uneth-

ical use of social media data can quickly and some-

times even quite unexpectedly impact the data access

policy decisions of social media platforms. However,

based on table 1 and table 2, we can see that there

is a long-term trend related to particularly limiting or

removing the access to explicit relationship data and

metadata from social media. We ﬁnd that tie strength

researchers should be prepared to even further signif-

icant changes in data policies, which are impacting

tie strength research. Tie strength researchers should

make themselves quite aware of the risks of such po-

tential changes to their tie strength research.

Third, there is a strong need to shift focus in tie

strength research to such approaches that can enable

the use of currently available data sources and data

items. Tie strength researchers should focus more

on using methods that are mainly or solely based on

available social media discussions and content. For

example, instead of explicit relationship social media

data, researchers can use implicitly deduced ties (see

e.g., Gupta et al. (2016)) for developing new social tie

strength models.

Fourth, tie strength researchers should look at the

existing wide variety and different kind of theoretical

approaches and other approaches that have been used

in other research areas or with other kinds of social

data (e.g., Mattie et al. (2018))and may be suitable for

social tie strength research. For example, such meth-

ods could include different kinds of existing and re-

cently built triadic closure models (see Huang et al.

Revisiting Social Media Tie Strength in the Era of Data Access Restrictions

193

(2015)) that help to make use of implicitly deduced

or predicted and potential social ties.

Finally, researchers should try to identify new le-

gal, sustainable, and innovative ways to access social

media data. These include buying social media data,

collaboration with other researchers in gaining nec-

essary social media data, and also combining addi-

tional data and other data sets to existing social media

data. For example, making better use of location and

co-location data and combining it with existing social

media data to make new social tie strength models.

6.2 Limitations and Future Research

This study has certain limitations. Firstly, this study

only focused on analyzing the changes to data ac-

cess policy of two social media platform - Facebook

and Twitter and not the other social media platforms.

Secondly, this study did not explain the exact techni-

cal details of the limitations introduced by the differ-

ent social media API versions(e.g., exact rate limits,

exact accessibility of different data items). Finally,

this study did not analyze how the new social media

platform feature changes (e.g., increase in the Tweet

length from 140 to 280 characters) can impact the cur-

rent social tie strength models.

This study leaves room for future studies. First,

in the future, a similar study could be done for some

other popular social media platforms Second, a study

about how the current social tie strength models are

adapted to the current data access situation. Finally,

do new studies to develop tie strength models which

can deduce social relationships(implicit relationship)

using just publicly available social media comments

and discussions.

REFERENCES

Burgess, J. and Bruns, A. (2012). Twitter archives and the

challenges of “big social data” for media and commu-

nication research. M/C J., 15.

Burgess, J. and Puschmann, C. (2014). Twitter and Soci-

ety: The Politics of Twitter. In Weller, K., Bruns, A.,

Marth, M., and Puschmann, C., editors, Twitter and

society, volume 89, chapter 4, pages xxix–xxxviii. Pe-

ter Lang Inc.

De Meo, P., Ferrara, E., Fiumara, G., and Provetti, A.

(2014). On Facebook, most ties are weak. Commu-

nications of the ACM, 57(11):78–84.

Fogues, R. L., Such, J. M., Espinosa, A., and Garcia-Fornes,

A. (2018). Tie and tag: A study of tie strength and tags

for photo sharing. PLoS ONE, 13(8):e0202540.

Gilbert, E. (2012). Predicting tie strength in a new medium.

In CSCW ’12, page 1047. ACM.

Gilbert, E. and Karahalios, K. (2009). Predicting tie

strength with social media. In CHI 09, page 211.

ACM.

Granovetter, M. S. (1973). The Strength of Weak Ties.

American Journal of Sociology, 78(6):1360–1380.

Gupta, J. P., Menon, K., K

arkk

ainen, H., Huhtam

aki, J.,

Mukkamala, R. R., Hussain, A., Vatrapu, R., Jussila,

J., and Pirkkalainen, H. (2016). Identifying weak ties

from publicly available social media data in an event.

In AcademicMindtrek ’16, pages 11–19. ACM.

Hogan, B. (2018). Social media giveth, social media taketh

away: Facebook, friendships, and APIs. International

Journal of Communication, 12(0):592–611.

Huang, H., Tang, J., Liu, L., Luo, J., and Fu, X. (2015). Tri-

adic Closure Pattern Analysis and Prediction in Social

Networks. IEEE TKDE, 27(12):3374–3389.

Huhtam

aki, J. and Olsson, T. (2018). Digital Social Match-

ing Ecosystem for Knowledge Work. In KMIS 10,

pages 194–199. SCITEPRESS.

Jones, J. J., Settle, J. E., Bond, R. M., Fariss, C. J., Mar-

low, C., and Fowler, J. H. (2013). Inferring Tie

Strength from Online Directed Behavior. PLoS ONE,

8(1):e52168.

Kahanda, I. and Neville, J. (2009). Using transactional in-

formation to predict link strength in online social net-

works. ICWSM09, pages 74–81.

Liberatore, F. and Quijano-Sanchez, L. (2017). What do we

really need to compute the Tie Strength? An empirical

study applied to Social Networks. Comput Commun,

110:59–74.

Mattie, H., Engø-Monsen, K., Ling, R., and Onnela, J.-P.

(2018). Understanding tie strength in social networks

using a local “bow tie” framework. Scientiﬁc Reports,

8(1):9349.

Onnela, J.-P., Saram

aki, J., Hyv

onen, J., Szab

o, G., Lazer,

D., Kaski, K., Kert

esz, J., and Barab

asi, A.-L. (2007).

Structure and tie strengths in mobile communication

networks. PNAS, 104(18):7332–6.

Quijano-S

anchez, L., D

ıaz-Agudo, B., and Recio-Garc

ıa,

J. A. (2014). Development of a group recommender

application in a Social Network. Knowledge-Based

Systems, 71:72–85.

Xiang, R., Neville, J., and Rogati, M. (2010). Modeling re-

lationship strength in online social networks. In WWW

’10, page 981.

Zhang, B., K

arkk

ainen, H., Gupta, J. P., and Menon, K.

(2017). The role of weak ties in enhancing knowl-

edge work. In AcademicMindtrek ’17, pages 210–215.

ACM.

KMIS 2019 - 11th International Conference on Knowledge Management and Information Systems

194