Face­book: Strug­gling with hate speech

Nu­dity, vi­o­lence and ter­ror­ist pro­pa­ganda eas­ier to tar­get

Albuquerque Journal - - NATION & WORLD - AS­SO­CI­ATED PRESS

SAN FRAN­CISCO — Get­ting rid of racist, sex­ist and other hate­ful re­marks on Face­book is more chal­leng­ing than weed­ing out other types of un­ac­cept­able posts be­cause com­puter pro­grams still stum­ble over the nu­ances of hu­man lan­guage, the com­pany re­vealed Tues­day.

Face­book’s self-as­sess­ment showed its polic­ing sys­tem is far bet­ter at scrub­bing graphic vi­o­lence, gra­tu­itous nu­dity and ter­ror­ist pro­pa­ganda. Au­to­mated tools de­tected 86 per­cent to 99.5 per­cent of the vi­o­la­tions Face­book iden­ti­fied in those cat­e­gories.

For hate speech, Face­book’s hu­man re­view­ers and com­puter al­go­rithms iden­ti­fied just 38 per­cent of the vi­o­la­tions. The rest came af­ter Face­book users flagged the of­fend­ing con­tent for re­view.

Face­book also dis­closed that it dis­abled nearly 1.3 bil­lion fake ac­counts in the six months end­ing in March. Had the com­pany failed to do so, its user base would have swelled be­yond its cur­rent 2.2 bil­lion. Fake ac­counts have got­ten more at­ten­tion in re­cent months af­ter it was re­vealed that Rus­sian agents used them to buy ads to try to in­flu­ence the 2016 elec­tions.

Even af­ter all that dis­abling, though, Face­book has said that 3 per­cent to 4 per­cent of its ac­tive monthly users are fake, mean­ing up to 88 mil­lion fake ac­counts slip through.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.