=Paper= {{Paper |id=Vol-1996/paper7 |storemode=property |title=Recognizing Images of Eating Disorders in Social Media |pdfUrl=https://ceur-ws.org/Vol-1996/paper7.pdf |volume=Vol-1996 |authors=Robert Pless,Rebecca Begtrup,Lulwah Alkulaib,Samsara N. Counts,James Harnett,Justine-Louise Manning,Hong Xuan,David A. Broniatowski |dblpUrl=https://dblp.org/rec/conf/amia/PlessBACHMXB17 }} ==Recognizing Images of Eating Disorders in Social Media== https://ceur-ws.org/Vol-1996/paper7.pdf 
                                Recognizing​ ​Images​ ​of​ ​Eating​ ​Disorders​ ​in​ ​Social​ ​Media​ ​(Abstract)

       Robert​ ​Pless,​ ​PhD​1​,​ ​Rebecca​ ​Begtrup,​ ​DO​ ​MPH​1,2​,​ ​Lulwah​ ​Alkulaib​1​,​ ​Samsara​ ​N.​ ​Counts​1​,​ ​ ​James​ ​Harnett​1​,
                               Justine-Louise​ ​Manning​1​,​ ​Hong​ ​Xuan​1​,​ ​David​ ​A.​ ​Broniatowski,​ ​PhD​1
   1​
      George​ ​Washington​ ​University,​ ​Washington,​ ​DC,​ ​USA;​ 2​​ Children’s​ ​National​ ​Health​ ​System,​ ​Washington,​ ​DC,
                                                                        USA

Eating disorders (ED) are pervasive and do not discriminate based on race, religion, gender, or SES. Comorbidities
include anxiety, depression, substance abuse, self-injurious behaviors, and history of trauma. ED are often a lifelong
struggle​ ​with​ ​approximately​ ​⅔​ ​of​ ​patients​ ​never​ ​achieving​ ​a​ ​full​ ​and​ ​sustained​ ​remission.
            ED are the product, in part, of increased societal pressures to fit "the thin ideal". These pressures come in
the form of repeated advertisements on various media platforms, messages from the diet and exercise industries,
fashion industry "norms", etc. Individuals who suffer from ED may have experienced trauma and/or have difficult
home​ ​lives.​ ​The​ ​ED​ ​can​ ​provide​ ​a​ ​sense​ ​of​ ​control​ ​over​ ​these​ ​factors,​ ​albeit​ ​an​ ​invalid​ ​one.
            Exposure to media expressing “the thin ideal” can be triggering to individuals with ED as well as those at
risk for developing them. Social media platforms are especially rife with these triggers. Concurrent with the rise of
social media, individuals with ED have created communities​1 in which they support one another in the dangerous
pursuit of this illness' elusive goal: to be “thin enough”. Websites promoting anorexia (pro-ana) and bulimia
(pro-mia) as lifestyle choices valorize acting on ED symptoms. Such sites teach those suffering or at risk from ED
how to develop, act on, and hide the illness, and support them in doing so, putting them at risk for serious physical
and​ ​mental​ ​health​ ​complications,​ ​including​ ​death.
            The impact of images in this community far exceeds that of other communities surrounding physical and
mental health issues. Therefore, it is important that clinicians and family members be able to identify websites
containing images that are associated with promotion of anorexia and bulimia in order to prevent accidental or
intentional exposure to these triggers. This research aims to automatically identify such triggering material, with the
ultimate​ ​goal​ ​of​ ​designing​ ​parental​ ​and​ ​clinical​ ​controls.
            We report on a proof of concept, machine learning approach to identify pro-ana content, trained on example
data from online social media searches. The training data is chosen to compare pro-ana content with other content
similar​ ​in​ ​demographics​ ​and​ ​photographic​ ​style:
   ● “pro-ana”: 16,000 images from a collection of Tumblr blogs including
     best-thinspo​,​ ​thinniest​,​ ​and​ ​wanna-be-skinnyminnie​.
   ● “selfie”:​ ​4,500​ ​Tumblr​ ​images​ ​tagged​ ​“selfie”.
   ● “ootd”:​ ​7,000​ ​Tumblr​ ​images​ ​tagged​ ​“ootd”​ ​(outfit​ ​of​ ​the​ ​day).
   ● “Greek”:​ ​5,000​ ​images​ ​from​ ​Tumblrs​ ​of​ ​Greek-letter​ ​college​ ​organizations.
We randomly choose 4740 (15%) of these images to reserve test data, and train
the Resnet Deep Learning neural network​2 to classify the remaining training
images into these categories.                 On test data this gives 78% classification
accuracy—a significant improvement over chance (25%). To explore a possible application, we identify 10
additional tumblr accounts, five that we judged to have high pro-ana content, 4 blogs without pro-ana content, and 1
fitness inspiration (fitspo) blog that we judged to contain a mix of content. The table below shows the percentage of
images​ ​classified​ ​as​ ​pro-ana​ ​in​ ​each​ ​blog:
​ ​ ​ ​ ​Blog​ ​Type​ ​ ​Title,​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​%pro-ana​ ​ ​ ​ ​ ​Blog​ ​type​ ​ ​ ​ ​Title​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​%pro-ana
 ​ ​ ​ ​ ​pro-ana​ ​ ​ ​ ​think-thygap​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​73​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​not​ ​pro-ana​ ​ ​abelmvada​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​7
  ​ ​ ​ ​ ​pro-ana​ ​ ​ ​ ​thinninglittle​ ​ ​ ​ ​ ​ ​ ​ ​ ​88​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​not​ ​pro-ana​ ​ ​roommysocks​ ​ ​ ​ ​ ​ ​ ​ ​ ​21
   ​ ​ ​ ​ ​pro-ana​ ​ ​ ​ ​think-skinny-th0ughts​ ​ ​89​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​not​ ​pro-ana​ ​ ​mathematicalmemer​ ​ ​ ​ ​4
    ​ ​ ​ ​ ​pro-ana​ ​ ​ ​ ​oh2beskinny​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​83​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​not​ ​pro-ana​ ​ ​satoshikurosaki​ ​ ​ ​ ​ ​10
     ​ ​ ​ ​ ​fitspo​ ​ ​ ​ ​ ​veganpilatesangel​ ​ ​ ​ ​ ​ ​53​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​not​ ​pro-ana​ ​ ​traitspourtraits​ ​ ​ ​ ​ ​9

These​ ​proof​ ​of​ ​concept​ ​results​ ​suggest​ ​that​ ​it​ ​is​ ​feasible​ ​to​ ​automatically​ ​detect​ ​social​ ​media​ ​sources​ ​with​ ​triggering
material,​ ​informing​ ​the​ ​creation​ ​of​ ​tools​ ​that​ ​can​ ​assist​ ​clinicians​ ​and​ ​family​ ​members​ ​to​ ​improve​ ​health​ ​outcomes.
1. O​ksanen​ ​A,​ ​Garcia​ ​D,​ ​Räsänen​ ​P.​ ​Proanorexia​ ​communities​ ​on​ ​social​ ​media.​ ​Pediatrics.​ ​2016​ ​Jan​ ​1;137(1):e20153372.
2. He​ ​K,​ ​Zhang​ ​X,​ ​Ren​ ​S,​ ​Sun​ ​J.​ ​Deep​ ​residual​ ​learning​ ​for​ ​image​ ​recognition.​ ​In​ ​Proceedings​ ​of​ ​the​ ​IEEE​ ​Conference​ ​on
   Computer​ ​Vision​ ​and​ ​Pattern​ ​Recognition​ ​2016​ ​(pp.​ ​770-778).