In­sta­gram posts may hold clues to a user’s men­tal health

The Globe and Mail (Prairie Edition) - - GLOBE LIFE & ARTS - NIRAJ CHOKSHI

The pho­tos you share on­line speak vol­umes. They can serve as a form of self-ex­pres­sion or a record of travel. They can re­flect your style and your quirks. But they might con­vey even more than you re­al­ize: The pho­tos you share may hold clues to your men­tal health, new re­search sug­gests.

From the colours and faces in their pho­tos to the en­hance­ments they make be­fore post­ing them, In­sta­gram users with a his­tory of de­pres­sion seem to present the world dif­fer­ently from their peers, ac­cord­ing to the study, pub­lished re­cently in the jour­nal EPJ Data Science.

“Peo­ple in our sam­ple who were de­pressed tended to post pho­tos that, on a pixel-by-pixel ba­sis, were bluer, darker and greyer on av­er­age than healthy peo­ple,” said An­drew Reece, a post­doc­toral re­searcher at Har­vard Uni­ver­sity and co-au­thor of the study with Christo­pher Dan­forth, a pro­fes­sor at the Uni­ver­sity of Ver­mont.

The pair iden­ti­fied par­tic­i­pants as “de­pressed” or “healthy” based on whether they re­ported hav­ing re­ceived a clin­i­cal di­ag­no­sis of de­pres­sion in the past. They then used ma­chine-learn­ing tools to find pat­terns in the pho­tos and to cre­ate a model pre­dict­ing de­pres­sion by the posts.

They found that de­pressed par­tic­i­pants used fewer In­sta­gram fil­ters, those which al­low users to dig­i­tally al­ter a photo’s bright­ness and colour­ing be­fore it is posted. When these users did add a fil­ter, they tended to choose “Inkwell,” which drains a photo of its colour, mak­ing it black-and­white. The health­ier users tended to pre­fer “Va­len­cia,” which light­ens a photo’s tint.

De­pressed par­tic­i­pants were more likely to post pho­tos con- tain­ing a face. But when health­ier par­tic­i­pants did post pho­tos with faces, theirs tended to fea­ture more of them, on av­er­age.

As re­veal­ing as the find­ings are about In­sta­gram posts specif­i­cally, both Reece and Dan­forth said the re­sults speak more to the prom­ise of their tech­niques.

“This is only a few hun­dred peo­ple, and they’re pretty spe­cial,” Dan­forth said of the study par­tic­i­pants. “There’s a sieve we sent them through.”

To be in­cluded in the study, par­tic­i­pants had to meet sev­eral cri­te­ria. They had to be ac­tive and highly rated on Ama­zon’s Me­chan­i­cal Turk plat­form, a paid crowd­sourc­ing plat­form that re­searchers of­ten use to find par­tic­i­pants. They also had to be ac­tive on In­sta­gram and will­ing to share their en­tire post­ing his­tory with the re­searchers. Fi­nally, they had to share whether or not they had re­ceived a clin­i­cal di­ag­no­sis of de­pres­sion.

Out of the hun­dreds of re­sponses they re­ceived, Reece and Dan­forth re­cruited a to­tal of 166 peo­ple, 71 of whom had a his­tory of de­pres­sion. They col­lected nearly 44,000 pho­tos in all.

The re­searchers then used soft­ware to an­a­lyze each photo’s hue, colour sat­u­ra­tion and bright­ness, as well as the num­ber of faces it con­tained. They also col­lected in­for­ma­tion about the num­ber of posts per user and the num­ber of com­ments and likes on each post.

Us­ing ma­chine-learn­ing tools, Reece and Dan­forth found that the more com­ments a post re­ceived, the more likely it was to have been posted by a de­pressed par­tic­i­pant. The op­po­site was true for likes. And de­pressed users tended to post more fre­quently, they found.

Though they warned that their find­ings may not ap­ply to all In­sta­gram users, Reece and Dan­forth ar­gued that the re­sults sug­gest that a sim­i­lar ma­chine­learn­ing model could some day prove use­ful in con­duct­ing or aug­ment­ing men­tal-health screen­ings.

“We re­veal a great deal about our be­hav­iour with our ac­tiv­i­ties,” Dan­forth said, “and we’re a lot more pre­dictable than we’d like to think.”


Ac­cord­ing to a study, users with a his­tory of de­pres­sion can be iden­ti­fied through clues such as colours and fil­ters.

Newspapers in English

Newspapers from Canada

© PressReader. All rights reserved.