The HR Digest - - Content Features -

Sen­ti­ment Anal­y­sis, the new hero in town!

Sen­ti­ment-anal­y­sis soft­ware can help com­pa­nies fig­ure out what’s both­er­ing work­ers—or what they’re ex­cited about. The term ‘sen­ti­ment anal­y­sis,’ is self­ex­plana­tory, but for the sake of it, we’ll list a def­i­ni­tion we found on Google - “the process of com­pu­ta­tion­ally iden­ti­fy­ing and cat­e­go­riz­ing opin­ions ex­pressed in a piece of text, es­pe­cially in or­der to de­ter­mine whether the writer’s at­ti­tude to­wards a par­tic­u­lar topic, prod­uct, etc. is pos­i­tive, neg­a­tive, or neu­tral.”

Every day, hu­mans, col­lec­tively, type out over 200 bil­lion emails, 500 mil­lion of tweets, and hun­dreds of mil­lions of texts, chats, and pri­vate mes­sages. It’s prac­ti­cally im­pos­si­ble for one sin­gle per­son to stitch to­gether these data, to fig­ure out emo­tional trends or be­hav­ioral themes. This is why we have com­put­ers. For decades, re­searchers have been de­vel­op­ing com­puter pro­grams which can try and un­der­stand the emo­tions stirred up by an idea or a prod­uct from our writ­ing. This par­tic­u­lar field, is known as ‘sen­ti­ment anal­y­sis.’ As a mat­ter of fact, it’s pretty pop­u­lar in the world of mar­ket­ing, and is com­monly re­ferred to as ‘opin­ion min­ing’. It refers to the anal­y­sis of one’s feel­ing (i.e. emo­tions, opin­ions, and at­ti­tudes) be­hind the big blur of words us­ing lan­guage pro­cess­ing tools. The idea is to use com­put­ers to look be­yond the veneer of con­strued words – are you pos­i­tive, sar­cas­tic, neg­a­tive, or bi­ased?

So, what does sen­ti­ment anal­y­sis have to do with em­ploy­ers and their busi­ness? You see, it all goes back to the mid2000s, when com­pa­nies wanted to un­der­stand how peo­ple re­spond to their prod­ucts, or their com­peti­tors of­fer­ings. Al­go­rithms were be­ing used to ag­gre­gate re­views to re­veal broader in­sights than sur­veys or fo­cus groups. This grew to the point that dozens of star­tups are now ex­clu­sively of­fer­ing these sen­ti­ment anal­y­sis soft­wares to let them know how their own em­ploy­ees feel about their jobs.

Large cor­po­ra­tions like IBM, Twit­ter, In­tel, and Ac­cen­ture have started in­sti­tut­ing the soft­ware to un­der­stand how their work­ers feel about their jobs. The aim here is to iden­tify prob­lems that might eas­ily es­cape a su­per­vi­sor dur­ing the an­nual per­for­mance re­view.

Ear­lier this year, IBM started us­ing sen­ti­ment anal­y­sis soft­ware to bet­ter re­tain em­ploy­ees in the com­pet­i­tive job mar­ket. The soft­ware uses lan­guage­pro­cess­ing and ma­chine-learn­ing

al­go­rithms to de­ci­pher emo­tions from text found in open-ended ques­tions on com­pany sur­veys, com­ments on com­pany blogs, and in­ter­nal so­cial net­work­ing sites.

In­tel uses a sim­i­lar soft­ware from Kan­joya Inc., to bet­ter un­der­stand em­ployee frus­tra­tion. The soft­ware turned out to be pretty in­sight­ful, as well. It re­vealed how a ma­jor­ity of the em­ploy­ees had the wrong im­pres­sion that their own jobs were at risk.

Such cases high­light how im­per­a­tive is it for com­pa­nies to rely on such tech­nolo­gies. Ac­cord­ing to CIO Jour­nal colum­nist Irv­ing Wladawsky-berger, in to­day’s world, where work­place col­lab­o­ra­tion is the key and where tech­nol­ogy-em­pow­ered cus­tomers can eas­ily share what they think about a prod­uct or ser­vice, em­pa­thy is the com­pet­i­tive edge.

Not so long ago, Twit­ter hired Kan­joya, to an­a­lyze work­ers’ re­sponses to com­pany sur­veys about their work­place ex­pe­ri­ences. The sur­veys were ad­min­is­tered twice yearly, and in­cluded only two open-ended ques­tions. Af­ter hir­ing Kan­joya, Twit­ter started send­ing the sur­vey to one-sixth of its work­ers every month – it also in­creased the num­ber of open-ended ques­tions. The pat­terns ex­tracted from Kan­joya’s anal­y­sis plat­form were then shared with the ex­ec­u­tives.

Kan­joya, also ad­ver­tises that its sen­ti­ment anal­y­sis tools work with Yam­mer, a so­cial net­work ac­quired by Mi­crosoft for a bil­lion dol­lars. Some of Kan­joya’s prod­uct of­fer­ings in­clude em­ployee en­gage­ment track­ing (to trace pos­i­tive or neg­a­tive emo­tions), and a search func­tion which re­sponds to queries which an anal­y­sis of the sur­round­ing sen­ti­ment. A lot of com­pa­nies to­day are more fo­cused on an­a­lyz­ing em­ployee chat­ter out­side of the for­mal re­views or sur­veys. Now, this makes it dif­fi­cult to scoop a struc­tured re­sponse, or iden­tify be­hav­ioral themes. IBM has for years, been col­lect­ing em­ploy­ees’ posts and com­ments on its in­ter­nal so­cial net­work­ing plat­form.

Called Con­nec­tions, it’s avail­able to all of IBM’S 400,000 em­ploy­ees world­wide. It func­tions like a mélange of Face­book, Wikipedia, and Dropbox, al­low­ing em­ploy­ees to pub­lish posts, com­ment on oth­ers’, and col­lab­o­rate with one an­other on cer­tain projects. IBM also sells a ver­sion of the plat­form, Con­nec­tions to other com­pa­nies. The so­cial net­work­ing plat­form is in­te­grated with a sen­ti­ment anal­y­sis tool called So­cial Pulse, which al­lows IBM to mon­i­tor posts and com­ments for be­hav­ioral trends and red flags.

In 2015, IBM used So­cial Pulse to re­vamp its per­for­mance-re­view sys­tem. Its HR depart­ment re­lapsed the old feed­back sys­tem to cre­ate a new one in or­der to re­ceive gen­uine re­sponses. IBM used So­cial Pulse to comb through the hun­dreds of thou­sands of feed­back re­ceived.

The soft­ware nar­rated an en­tirely new story: Em­ploy­ees at IBM were un­happy that their per­for­mances were ac­tu­ally graded on a curve. Within a month or two, the com­pany in­tro­duced a new and im­proved method.

By widen­ing the scope of the data ac­cu­mu­lated via sur­veys, re­views, and so­cial me­dia posts, there’s a cer­tain risk of vi­o­lat­ing em­ploy­ees’ pri­vacy. This is why IBM lim­its the data-min­ing to posts and com­ments that are shared with the en­tire com­pany. It bars emails, chats, or in­ter­ac­tions in pri­vate groups.

One can­not en­tirely rely on sen­ti­ment anal­y­sis for a pol­ished re­port. You see, the com­puter’s un­der­stand­ing of nat­u­ral lan­guage, is still in its in­fancy. Ac­cord­ing to a re­search project (2016), ba­sic anal­y­sis tools sent be­tween de­vel­op­ers and an open-source server soft­ware suite only had a max­i­mum ac­cu­racy rate of 30 per­cent. How­ever, when two peo­ple tried to de­ter­mine the emo­tions ex­pressed in 50 emails, they could only agree on three-quar­ters of them. These al­go­rithms still lack the hu­man el­e­ments, em­pa­thy. A small team of an­a­lysts rou­tinely ex­am­ine IBM’S So­cial Pulse, to en­sure they’re send­ing the right trends to the man­age­ment.

We’re still a long way to go, if we re­ally want to im­prove our abil­ity to un­der­stand how our em­ploy­ees feel. A group of com­puter sci­en­tists in In­dia pub­lished a pa­per which sug­gests a new way of de­ter­min­ing em­ploy­ees’ well-be­ing: fa­cial scans. The sys­tem uti­lizes images cap­tured of em­ploy­ees’ faces each time they en­ter the build­ing, to de­ter­mine whether they’re happy, de­pressed, sad, or an­gry. The fa­cial scans might some­day help busi­ness use the date to op­ti­mize pro­duc­tiv­ity and prof­its. Un­til then, there are plenty of emo­tion-de­tec­tion tech­nolo­gies in the mar­ket, and to be hon­est, it’s not a bad start.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.