Semantic Networks of Interests in Online NSSI
Communities
A Case Study in Mining Pathological Adolescent Minds
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Mathematics and Computer Science Department, Psychology Department
Suffolk University, Boston
June 22, 2012

What Is NSSI?
Non-suicidal self-injury (NSSI) is the direct, deliberate destruction of
one’s own body tissue in the absence of suicidal intent.
It is practiced primarily by adolescents and young adults and is often
concealed from others.
Common NSSI activities include skin cutting, banging or hitting
oneself, and burns.
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

Why NSSI Matters?
14% to 21% of adolescents and 17% to 25% of young adults have
engaged in NSSI at some point in their lives
NSSI is repeatedly found to be associated with significant emotional
and behavioral dysfunction (e.g., eating disorders, suicide).
Can we identify NSSI persons by automatically analyzing secondary
data publicly available from massive online social networks (MOSN)?
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

NSSI and Massive Online Social Networks
Many popular MOSNs (e.g., Facebook and LiveJournal) allow users
to declare their interests.
Associations between interest lists and NSSI community membership
suggest that “likes” or interest lists may be serving as identity
signals.
Profile page with interests of a random NSSI LiveJournal user:
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

NSSI Communities in LiveJournal
LiveJournal[.com] is a popular massive online blogging social
network site (BSN)—a bimodal venue where users engage in both
publishing and social activities.
Bloggers can form contact lists, subscribe to their friends’ blogs,
comment on selected blog posts, declare interests, and participate in
communities—collective blogs.
32 mln individual and community accounts.
43 identified NSSI-related communities (some of them promote
NSSI activities, while others advocate for NSSI abstinence) with
22,000 members and/or posters.
The total number of harvested interests is ∼150,000, including
misspelled, abbreviated, and hyphenated variants.
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

Users and Interests
M is the adjacency matrix of a two-mode network of the NSSI users and
their interests:
Mij =
1
0
iff user Ui declares interest Vj
else
Should we use Pearson correlation (a.k.a. cosine distance) to calculate
similarities of interests?
Yes—if all users were different. However, a thematic community is likely
to be more homogeneous than a random group of users.
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

Kova´z Correlations
c
B. Kova´z: “Two terms are similar with correlation Θij ∈ [−1, 1] if they
c
are used by similar people; two people are similar with correlation
Φij ∈ [−1, 1] if they use similar terms.” (We calculate Φ but do not use
it.)
Initialize:
Mi = Mi, − Mi,
Mj = M
,j
−M
,j
Θij,0 = Φij,0 = δij
Iterate:
Θij,k+1 = Mi Φk MjT /
Mi Φk MiT (M j Φk M jT )
Φij,k+1 = MiT Θk Mj /
MiT Θk Mi (M jT Θk M j )
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

From Θ to an Adjacency Matrix
Θ is a dense symmetric signed square matrix with few or no zero
terms.
The distribution of Θij is close to uniform.
Restricted to the top 600 most often declared interests (due to
computational resources).
Convert Θ to a sparse adjacency matrix Ψ:
Ψij =
Θij
0
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
if Θij ≥ 0.8
else
Semantic Networks of Interests in Online NSSI Communities

Semantic Map, as Seen by Gephi
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

Clustering
Four major clusters: “music” (MUS), “pathology” (PAT), “daily
life/emotions” (DLE), and “creativity” (CRE). Examples of terms:
MUS: atreyu, him, incubus, korn, my chemical romance, nirvana,
rancid, system of a down, the perfect circle;
PAT: alcohol, anorexia, bulimia, burning, cutting, handcuffs,
pain, self-injury and self-mutilation (both with and
without the dash), spikes, weeds;
DLE: cameras, cloths, dvds, flirting, flowers, fun, quotes,
smiling, hearts (also as an HTML entity ♥ and as
♥);
CRE: astrology, books, languages, philosophy, psychology,
shakespeare, sociology, travel, wine.
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

Bridges
The border zones are spanned with few bridge interests:
PAT/MUS: (black) eyeliner, girl interrupted, metal;
PAT/DLE: candy, girls, insomnia, red, rock music, sex;
MUS/DLE: animals, camping, fashion, games, honesty, humor,
travel(l)ing;
All clusters: bands, bracelets, hoodies, lesbians, making out.
Can these bridges be viewed as NSSI “beacons?”
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

Pathology Cluster
The blowout of the PAT cluster:
(A link between two interests means many people often mention these
interests together.)
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

Are the Links NSSI Specific?
Compare the NSSI communities to a random sample of LiveJournal
users—doesn’t work, they share very few interests!
Find appropriate communities that may share interests with the
NSSI population:
SMM “sexy-mood-music” (6,700 members, average age 25
years, music),
M15M “movies-in-fifteen-minutes” (13,300 members,
average age 28 years, video fans).
Calculate the intersection between the NSSI semantic network and
each of the other semantic networks under consideration. The
intersection contains the associations that are significant for both
communities and presumably are pathology-free.
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

Pathology-Free Networks
Intersection of the NSSI and SMM networks:
Well-defined CRE and DLE clusters. The MUS cluster is sparse and
subdivided into movies and proper music. The PAT cluster is gone!
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

Leisure on Their Mind
Our findings appear indicative of the growing global middle-class youth
culture revolving around leisure activities (e.g., music, art) reflecting
adolescent development in internationally-connected networks. This is
supported by the similarities between the NSSI interested communities
and the non-pathological comparison communities.
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

Conclusions
We constructed a semantic network of interests declared by
non-suicidal self-injury (NSSI) bloggers of LiveJournal.
The network consists of four clearly separated interest clusters
corresponding to the pathological terms, daily life, popular music,
and creativity.
The interests that bridge gaps between the pathology cluster and
the other three clusters can be used as beacons signaling the
potential presence of an NSSI behavior.
The extent of MOSN NSSI-related communities on LiveJournal
could evidence the limited opportunities for social networking among
people.
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities

Thanks!
This research has been supported in part by the College of Arts and
Sciences, Suffolk University, through an undergraduate research
assistantship grant.
The authors are grateful to Zo¨ Wells of Suffolk University for preliminary
e
data collection and Dr. Jim Hollander and Prof. John Boyd for
suggestions on combining graphs.
Dmitry Zinoviev, Dan Stefanescu, Lance Swenson, Gary Fireman
Semantic Networks of Interests in Online NSSI Communities