Hashtag of Instagram: From Folksonomy to Complex Network

Simona Ibba, Matteo Orrù, Filippo Eros Pani and Simone Porru

Department of Electrical and Electronic Engineering, University of Cagliari, Piazza d'Armi, Cagliari, Italy

Keywords: Instagram, Social Tagging, Knowledge Management, Complex Network, Folksonomy.

Abstract: The Instagram is a social network for smartphones created in 2010 and acquired by Facebook in 2012. It

currently has more than 300 million registered users and allows for the immediate upload of images (square,

inspired by Polaroid), to which users can associate hashtags and comments. Moreover, connections can be

created between users that share the same interests. In our work, we intend to analyze the hashtags entered

by users: the use of such hashtags, as it happens in other social networks like Twitter, generates a

folksonomy, that is a user-driven classification of information. We intend to map that folksonomy as a

complex network to which we can associate all the typical analysis and evaluations of such a mathematical

model. Our purpose is to use the resulting complex network as a marketing tool, in order to improve brand

or product awareness.

1 INTRODUCTION

Social Media Marketing has become essential for

every company: it is almost free, and, if used

correctly, it is incomparable in terms of audit and

Customer Relationship Management.

The main purpose of social strategy is to increase

brand reputation, brand loyalty and brand awareness.

The main tools of Social Media Strategy have seen

the recent entry in their ranks of Instagram, together

with the more widespread Twitter, Facebook or

Youtube. Instagram is a mobile application started in

2010, that in 2015 surpassed the mark of 300 million

registered users. Instagram has different features

from the other tools: it is an exclusively mobile

application, it does not allow the posting of direct

links to sites, and has a difficult integration with

Social Media Management tools. Those

characteristics, which could appear as limits, make

Instagram a very fitting tool for the creation of brand

awareness. The brand Nike, for example,

(http://instagram.com/nike) owns an Instagram

profile with 17,822,162 followers currently, with an

Avg Daily Followers number of 40,752

(http://socialblade.com/instagram/user/nike). These

figures would suggest that Instagram will soon

integrate advertising systems that would allow

brands to pay to reach users that are still not

followers. In the meantime, the main tool in the

hands of a manager of an Instagram profile is

hashtags: a suitable hashtag makes the content

visible to all users interested in that specific topic,

and also gives visibility to the profile, generating

additional followers. Good content with suitable

hashtags generate likes and comments that make the

account visible in the “Explore” tab, where new

content of interest can be found.

With our work, we intend to research the

hashtags entered by users, through the use of

Instagram's Application Programming Interface

(API). We plan to establish a relation between

hashtags and metadata associated to uploaded

images, and to analyze those relations. Social

tagging on Instagram leads to the generation of a

folksonomy, that is a collaborative, collective, and

social organization, at the metadata level, of

information entered by users, as suggested by

Angius et al., (2014). Lastly, our aim is to map

hashtags through a complex network and find a tool

for the creation of new hashtags relevant to the

images and that contribute to increase the visibility

of pictures.

The general objectives of our research are

summarized below:

 Analyze the properties of complex networks

originating from Instagram hashtags.

 Elaborate predictive models for the content

posted by users

 Locate the content posted by users on Instagram

to develop appropriate marketing strategies.

Ibba, S., Orrù, M., Pani, F. and Porru, S..

Hashtag of Instagram: From Folksonomy to Complex Network.

In Proceedings of the 7th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2015) - Volume 2: KEOD, pages 279-284

ISBN: 978-989-758-158-8

279

This paper is structured as follows: Section 2

describes the context in which the proposed project

lies, in Section 3 we present an overview about the

state of the art; in Section 4 we describe our

approach, while in Section 5 we explain our project

for the use of Instagram hashtags as a complex

network. The last section hosts our final

observations about the project.

2 BACKGROUND

Instagram allows the upload and sharing of photos

and videos through the use of a mobile device.

(a) (b)

Figure 1: Interfaces of Instagram. (a) Instagram photo with

comments (b) Hashtags entered by users.

It offers its users a unique way to post their

pictures, and allows immediate editing through 22

filters. It also allows users to add captions, hashtags

using the # symbol to describe photos or videos, or

to mention other users using the @ symbol (the @

symbol creates an actual link among the accounts).

Instagram also allows users to follow posts from all

selected profiles, and to have their own followers.

Each follower must be specifically approved by the

profile owner. Every user can set their own privacy

preferences, and can make the pictures visible to

everyone or only to followers. Pictures in profiles

are shown in chronological order, starting from the

latest. For each picture, it is possible to enter likes or

comments. Hashtags and user mentions can be

entered inside comments.

Moreover, according to a research by (Hu et al.,

2014), Instagram photos can be roughly categorized

into eight types based on their content: self-portraits,

friends, activities, captioned photos (pictures with

embedded text), food, gadgets, fashion, and pets,

where the first six types are much more popular.

3 RELATED WORK

In the literature, studies on Instagram are in

markedly smaller numbers than studies on Twitter.

The latest developments and the big figures in

Instagram usage as of late, lead us to believe that

this mobile app is deserving of more attention from

the scientific community. Understanding Instagram

and its mechanics means working on human,

cultural, social, and environmental dynamics that

open several scenarios, also from a business,

communication, and marketing perspective.

The main key to Instagram is the hashtag:

hashtags can function as descriptive elements in the

image, or can be related to its caption, explaining its

content better. Hashtags are thus a way to manage

and organize information, as stated by Mathes

(Mathes et al., 2014) as regards the social

bookmarking system in Delicious, in that it leads to

the generation of a folksonomy. Bruns and Burgess

(2014) maintain that the use of certain hashtags can

allow certain types of communities to emerge and

form, including ad hoc publics, forming and

responding very quickly in relation to a particular

event or topical issue.

The main purpose of social tagging is thus to

facilitate visibility of information (visibility of

images in Instagram's case) for the creation of

recommendation systems. Our work is based on the

intention of finding, in an automated way, using

complex networks and missing links, the most

suitable hashtags to the retrieval of images on

Instagram, making an account more visible as a

result. Similar works have been carried out by

Marlow et al., (2006), and Ames and Naaman

(2007), who view content management and retrieval

as two of the most important incentives to tag

resources. Hashtags in Instagram can describe the

content of the image, but can likewise represent

subjective opinions, feelings, places, or a variety of

expressions pertaining to colloquial language. In our

work, we took our cue from Bischoff et al., (2007),

who analyzed the data sets of tags extracted from

several tagging systems like Flickr and Delicious,

and studied the distribution of tags associated to the

different resources to find the implications derived

from the usage of different tags to improve research

and visibility of content. One of the first authors that

thought of building a network of tags and of using a

KEOD 2015 - 7th International Conference on Knowledge Engineering and Ontology Development

280

graph-based approach is Mika (2005). In their work,

they apply clustering techniques to tags, uses their

co-occurrence statistics, and produce conceptual

hierarchies. A useful approach to define sets of

related tags is the one devised by Gemmell et al.,

(2008): it is based on the hypothesis that clustering

tech1niques are better if the context in which the

tags appear can be determined. In De Gemmis et al.,

(2008) suggest an interpretation of tags through the

Wordnet thesaurus: related tags can be defined in

this way, reducing the issue of ambiguity.

The use of a folksonomy as a Complex Network

was introduced also by Shen et al., (2005), who

highlighted that, since folksonomy is a classification

system of web contents, its properties both static and

dynamic can also serve to search and retrieve the

related information.

Complex networks have recently received a lot

of attention by researchers of different disciplines

(Newman, 2003). These systems are found in many

area of the natural (metabolic network, etc) and

human (Internet, the web, power networks, etc).

Althogh their different nature, several different

networks share some properties: scale-free network,

small-world, (Valverde et al., 2002; 2003), and

community structure (Girva et al., 2001) (Fortunato,

2010), just to name a few of the most studied.

Many complex networks of different nature

present the small-world property (Milgram, 1967),

showing small values for mean shortest paths (if

compared to a random network of the same size) and

high value of the clustering coefficient. In this kind

of networks the paths that separate each pair of

nodes are relatively short.

Another interesting properties regards the

distribution of the degree (in and out) of complex

networks, that usually follows an exponential

distribution. This led to the definition of scale-free

networks. This kind of network present few nodes

having high degree, whereas the vast majority of

nodes have a low degree.

One of the most important property of complex

network is the community structure (Girvan et al.,

2001), (Fortunato, 2010). A community is a

subnetwork of densely connected nodes. The nodes

of a community are more connected to each other if

compared to nodes that are outside the community.

Determining a significant community structure is not

straight forward and many authors proposed their

algorithm to determine the community structure of a

network. One of the most used is the FastGreedy

algorithm (Clauset et al., 2004).

Complex networks show also a hierarchical

structure, namely the network's communities are

organized at different levels, being some of the

included into others. A method to retrieve the

hierarchical structure of a network has been

proposed by Clauset et al., (2007; 2008) and consist

in using a Monte-Carlo Markov Chain approach to

compute the dendrogram of an associated

hierarchical random graph. This method can be used

to for the detection of missing links between nodes.

Figure 2: Collaboration network of jazz musicians. Each

node is a Jazz musician and an edge denotes that two

musicians have played together in a band (Gleiser and

Danon, 2003). Different colors represent the community

retrived with the FastGreedy Algorithm (Clauset et al.

2004).

4 THE PROPOSED APPROACH

The proposed approach is based on the concept of

complex network (or graph), that is a structure with

two main separate elements: nodes, which represent

the basic elements of the graph, and node

connections, called branches or arches. An example

of a network we face every day is the road network.

We could associate cities to nodes, and roads

connecting cities to branches. The complexity is

intrinsic to the networks as their size grows, that is

to the number of involved elements (nodes and

branches), to the point that it becomes extremely

complex to understand its structure, its behavior, and

its evolution. For this reason, in order to study

complex networks, tools, methods, and algorithms

coming from a multi-field knowledge corpus

Hashtag of Instagram: From Folksonomy to Complex Network

281

(statistical physics, sociology, etc.) are used, a

corpus that has been gradually built over the years.

What makes complex networks interesting is the

fact that they constitute a mathematical model that

can represent facets and artifacts of human life, and

several natural phenomena. Networks of varied

origin and nature (from electric networks to

metabolic ones) possess the same properties.

It is possible to apply the same concept of

network to human knowledge. In this case, the term

knowledge network is used. An example of

knowledge network is represented by bibliographic

networks created from author collaboration (co-

authorship) information or quotations appeared in

reference sections.

In our work, we aim thus to map the hashtags

and metadata retrieved from the Instagram app as a

complex network, and we intend to study some

applications that, through the study of the network,

could be developed to make Instagram a more

rounded marketing tool.

In our study, we want to evaluate the effect of

hashtag categorization in the context of folksonomy-

based item recommendation using data crawled from

the multi-topic social networking system of

Instagram.

Shen k. et al. (Shen et al., 2005) demonstrated

that the folksonomy at Del.icio.us it can be studied

as a complex network formed by tags because

displays both nature of small world, scale free and

highly clustered.

We want to apply Instagram hashtags metrics

typical of complex networks. In particular we want

to study the incidence matrix of a subgraph to

determine the properties of the network. We can thus

identify the most important nodes, the ease of

transition from one node to another, the indices of

centrality, the hub nodes and those autority and

indexes of the graph. This information will allow us

to interpret the content posted by users and see

which are the most important information for each

geographical area and how this information can be

linked to content of other subgraphs network.

5 HASHTAG AS NODES OF A

COMPLEX NETWORK

Our analysis is based on the Instagram data collected

using the Instagram API. Instagram Application

Programming Interface (API) allows to research tags

placed by users, and provides complete information

on images and videos. Through the API, for each

photograph it is possible to find an univocal

identifier (id) and the link to two versions of the

same image (low resolution and standard resolution),

the metadata that describe user name, date and time

of image creation, the location where the picture was

taken, the caption entered by the author, comments,

tags associated to the image, number of likes and

names of the users that gave their like. Other than

the metadata associated to images, metadata

associated to each user that has posted a picture can

be extracted. These metadata allow to find number

of followers and following, email address, number

of posts and a brief biography.

The high number of available data allows to

perform quantitative and qualitative analysis, to

verify the stream of content over time and find the

most interesting topic to users. The information

gathered from it makes it also possible to map data

according to their geographic location, and to

analyze the geolocalization of users in relation to

specific tags of interest. The analysis on tags allows

to find users' consumption models, the type of

posted content, and the specific locations where the

same content is posted, together with time data.

As suggested in Laniado et al., (2010), who

analyzed Twitter's hashtags network, we are going to

define a set of metrics for describing hashtags usage

from different perspectives: frequency, specificity,

consistency and stability over time.

The main purpose of Instagram users, especially

of those users that manage communities

(http://instagramersitalia.it/iger/igers_sardegna/) is

to make their posted photos as visible as possible:

with a higher visibility, the number of followers

increases, and consequently so does the size of the

community and the number of users who can

automatically see the pictures posted by a specific

profile (and can thus comment or enter a like). The

pictures can represent a communication strategy, and

can be a part of a marketing plan. For example, they

could lead to the download of an app, or to the

promotion of a restaurant, etc. In order to make a

photo visible, and iteratively increase the number of

followers of the profile, it is necessary to use

suitable tags. Instagram users seldom find the most

suitable tags for a specific photo.

The aim of our work is, therefore, to map

Instagram tags through a complex network (or

complex graph). As proposed by (Cantador et al.,

2011) for the Flickr social network, we intend to use

an approach based on Random Walks with Restarts

theory (Lovasz, 1996), which allows us to directly

predict the preference of users to particular photos

from the data collection acquired, by taking into

KEOD 2015 - 7th International Conference on Knowledge Engineering and Ontology Development

282

account not only their personal profiles in terms of

item preferences but also their tagging behavior,

social network as well as similarly tagged items.

Specifically, we want to create our social graph

by representing users, photos and hashtags as nodes.

User relationships are encoded using either

unidirectional or bidirectional edges between the

corresponding nodes. Similarly, we add edges

between items and tags as well as users and

hashtags.

Through the analysis of the complex network, we

want to find the missing links, that is we want to

find and suggest new hashtags based on the first tags

entered by users that are found to be especially

suitable to the posted photo, consequently increasing

the visibility of the image: namely, socially relevant

tags. This analysis can be applied to the definition of

prediction algorithms that monitor sudden changes

in a network. This study is extremely interesting for

the search of trending topics associated to a specific

location or a specific user type.

6 CONCLUSION

Our proposal stems from the aim to analyze and use

the hashtags on Instagram. We hypothesized the

creation of a social graph, interpreting users, photos,

and hashtags as graph nodes. The relations between

these elements constitute the graph links:

unidirectional or bidirectional. We propose to

associate typical analysis of those models to the

complex network obtained with the above process.

We intend to interpret data gathered from it as useful

tools for marketing operations, so as to improve

brand awareness. In particular, we want to find

missing links to define new hashtags relevant to the

pictures uploaded on Instagram and to the profiles of

specific communities, so as to give a higher

visibility to profiles. In the future, we believe that

this analysis might be used for the interpretation of

the most relevant informative content for a specific

user type and in a specific location. In the tourism

industry, for example, the complex network and the

study of the missing links could provide, in a semi-

automated way, tour routes associated to different

user types that could be found though pictures

posted on Instagram. A user that visits a certain

region and posts pictures of its monuments could

automatically receive new suggestions of interesting

spots in their itinerary from the application.

REFERENCES

Ames, M., Naaman, M., 2007. Why we tag: motivations

for annotation in mobile and online media, in:

Proceedings of the 25th ACM Conference on Human

Factors in Computing Systems (CHI’07), 2007, pp.

971–980.

Angius A., Concas G., Manca D., Pani F. E., Sanna G.,

2014. Classification and indexing of web content

based on a model of semantic social bookmarking. In:

Proceedings of the 6th International Conference on

Knowledge Management and Information Sharing,

KMIS 2014, Rome, Italy, 21-24 October 2014. ISBN:

978-989-758-050-5

Bischoff, K., Firan, C.S., Nejdl, W. R., 2008. Can all tags

be used for search? In: Proceeding of the 17th ACM

Conference on Information and Knowledge

Management (CIKM’08), pp. 203–212.

Bruns, A., Burgess, J., 2011. “The use of Twitter hashtags

in the formation of ad hoc publics,” paper presented at

the European Consortium for Political Research

conference, Reykjavik (25–27 August), at

http://eprints.qut.edu.au/46515/, accessed 14 October

2014.

Cantador, I., Konstas, I., & Jose, J. M., 2011. Categorising

social tags to improve folksonomy-based

recommendations. Web Semantics: Science, Services

and Agents on the World Wide Web, 9(1), 1-15.

Clauset, A., Moore, C., Newman, M.E.J., 2008.

Hierarchical structure and the prediction of missing

links in networks. Nature 453, 98 - 101.

Clauset, A., Moore, C., Newman, M.E.J., 2007. Structural

Inference of Hierarchies in Networks. In E. M. Airoldi

et al. (Eds.): ICML 2006 Ws, Lecture Notes in

Computer Science 4503, 1 - 13. Springer-Verlag,

Berlin Heidelberg.

Clauset, A., Newman, M.E.J., Moore, C., 2004. Finding

community structure in very large networks. Phys.

Rev. E, 70(6):066111.

De Gemmis, M., Lops, P., Semeraro, G., Basile. P., 2008.

Integrating tags in a semantic content-based

recommender, in: Proceedings of the 2nd ACM

Conference on Recommender Systems (RecSys’08),

pp. 163–170.

Fortunato, S., 2010. Community detection in graphs.

Physics Report, 486:75–174.

Gemmell, J., Shepitsen, A, Mobasher, M., Burke, R.,

2008. Personalization in folksonomies based on tag

clustering, in: Proceedings of the 6th Workshop on

Intelligent Techniques for Web Personalization and

Recommender Systems.

Girvan, M., Newman, M. E. J., 2001. Community

structure in social and biological networks. Proc. Natl.

Acad. Sci. U. S. A., 99 (cond-mat/0112110):8271–

8276.

Gleiser P. M., Danon, L., 2003. Community structure in

jazz. Advances in Complex Systems, 6(4):565-573.

KONECT, 2015. Jazz musicians network dataset -

KONECT.

Hashtag of Instagram: From Folksonomy to Complex Network

283

Laniado, D., & Mika, P. (2010). Making sense of twitter.

In The Semantic Web–ISWC 2010 (pp. 470-485).

Springer Berlin Heidelberg.

Lovasz, l., 1996. Random walks on graphs: a survey,

Combinatronics 2, 1–46.

Marlow, C., Naaman, M., Boyd, D., Davis, M.,. 2006.

HT06, tagging paper, taxonomy, flickr, academic

article, toread, in: Proceedings of the 17th ACM

Conference on Hypertext and Hypermedia

(Hypertext’06), pp. 31–40.

Mathes, Adam, 2004. Folksonomies-cooperative

classification and communication through shared

metadata.

Mika, P., 2005. Flink: semantic web technology for the

extraction and analysis of social networks, Journal of

Web Semantics 3 (2–3) - 211–223.

Milgram, S.,1967. The small world problem, Psychology

Today 2, 60–67.

Newman, M. E. J., 2003. The structure and function of

complex networks. SIAM REVIEW, 45:167–256.

Shen, K., & Wu, L., 2005. Folksonomy as a complex

network. arXiv preprint cs/0509072.

Valverde S. Cancho, Sole R.V., Scale free networks from

optimal design. Europhysics Letters, 60, 2002.

Valverde S., Sole R.V., 2003. Hierarchical small worlds in

software architecture. arXiv:cond-mat/0307278v2.

KEOD 2015 - 7th International Conference on Knowledge Engineering and Ontology Development

284