Top 100 See - - Con­tents - By Nevena Krasteva

What is it you ac­tu­ally do and what dif­fer­ence do you make for a busi­ness or­gan­i­sa­tion?

Identrics op­er­ates in the area of text min­ing through ar­ti­fi­cial in­tel­li­gence. We de­velop tech­nolo­gies by which ma­chines can start to un­der­stand the nat­u­ral lan­guage of peo­ple so as to be able to cat­e­gorise various doc­u­ments. We work with nat­u­ral lan­guages, and ma­chines do not have a clue how nat­u­ral lan­guages work and they need to be trained in or­der to be able to find their way around. An­other thing we do is named en­ti­ties recog­ni­tion - we train a ma­chine to iden­tify a very spe­cific set of named en­ti­ties de­ter­mined by a client out of a data­base of hun­dreds of thou­sands of doc­u­ments. This is ac­tu­ally

one of our ma­jor com­pet­i­tive ad­van­tages to other com­pa­nies that of­fer “ready-made” en­tity recog­ni­tion and can­not meet a client's needs to sin­gle out a spe­cific le­gal or phys­i­cal per­son.

To­day, a num­ber of peo­ple em­ployed in the IT in­dus­try are trudg­ing through vast amounts of unin­spir­ing work on a daily ba­sis. This process could be au­to­mated, which would make it quicker and much cheaper, al­low­ing peo­ple to do more im­por­tant things which are still off lim­its for ma­chines, such as data anal­y­sis. Our mis­sion is to give peo­ple more room to be cre­ative and to tap the real po­ten­tial of cre­ative and as­so­cia­tive think­ing, thereby re­struc­tur­ing the cost of an in­for­ma­tion prod­uct and mak­ing more funds avail­able for cre­ative work.

What prac­ti­cal ap­pli­ca­tion do your so­lu­tions have?

Our tech­nolo­gies could be of use to any­one who is do­ing data re­search in­volv­ing an ex­ten­sive body of in­for­ma­tion. For ex­am­ple, if a re­cruit­ment agency whose main as­set is its pool of CVs has a client seek­ing to hire some­one for a very spe­cific po­si­tion, we can rel­a­tively quickly al­low the com­pany to find a suit­able can­di­date. To match some very spe­cific job re­quire­ments to a large pool of CVs, we only need to write the ideal can­di­date's CV, to in­dex it and then com­pare it to the body of info that we have.

An­other prac­ti­cal ap­pli­ca­tion of our so­lu­tions would be with com­pa­nies which have hun­dreds of thou­sands of cus­tomers and of­ten need cus­tomer sup­port cen­tres to an­swer their ques­tions, i.e. they of­ten have a rel­a­tively high cost for cus­tomer sup­port ser­vices. How­ever, cus­tomers of­ten ask iden­ti­cal or sim­i­lar ques­tions, and we could de­velop an al­go­rithm with the help of which as soon as a cus­tomer starts typ­ing a ques­tion, it would be matched to the body of al­ready an­swered ques­tions and a pos­si­ble an­swer of­fered.

How can you help me­dia or­gan­i­sa­tions in par­tic­u­lar?

Over the past decade me­dia in the re­gion have been in­vest­ing a lot in de­vel­op­ing at­trac­tive mul­ti­me­dia plat­forms that can be ac­ces­si­ble on various types of de­vice, and now these me­dia have state-of-the art dig­i­tal plat­forms via which they can dis­sem­i­nate in­for­ma­tion and serve ads. How­ever, the me­dia were slow to ac­knowl­edge that the in­for­ma­tion which they cre­ate is un­in­tel­li­gi­ble to ma­chines and that some ba­sic tasks such as the clas­si­fi­ca­tion of this in­for­ma­tion and the cre­ation of linked data can be done very quickly and with­out hu­man in­volve­ment. If an ar­ti­cle men­tions a politi­cian, this ar­ti­cle can au­to­mat­i­cally lead to a pro­file of this politi­cian, or to other ar­ti­cles with sim­i­lar con­tent, as the whole process of se­lec­tion of this linked data can be done by a ma­chine. If a me­dia plans to cre­ate a web­site that brings to­gether con­tent from al­ready ex­ist­ing web­sites, this too can be done fairly easy and quickly.

We of­fer the me­dia ac­cess to tech­nolo­gies which they would find it very hard to de­velop on their own. These tech­nolo­gies would rel­a­tively quickly im­prove ma­jor pa­ram­e­ters of the be­hav­ior of a me­dia's au­di­ence such as the du­ra­tion of read­ers' stay and their en­gage­ment. This is par­tic­u­larly im­por­tant for the me­dia in the post-cri­sis con­text in which they op­er­ate - where the en­try of heavy­weights such as Google and Face­book has changed fun­da­men­tally the ad mar­ket, leav­ing the me­dia with a very lim­ited fi­nan­cial re­source which they can in­vest in in­no­va­tions re­lated to ar­ti­fi­cial in­tel­li­gence. If a me­dia wants to di­ver­sify its busi­ness model and sell not only ads but some sort of re­search or analy­ses, it needs higher au­di­ence en­gage­ment.

What type of clients do you tar­get?

Our clients are com­pa­nies based in the US and the UK that have spe­cialised in data in­tel­li­gence and me­dia. How­ever, we would like to spe­cialise in the re­gion of South­east Europe, which means be­ing able to work with the lan­guages of the coun­tries in the re­gion. In fact, lan­guage it­self is one of the big­gest chal­lenges in com­pu­ta­tional lin­guis­tics.

Mak­ing sense of a lan­guage re­quires a huge in­vest­ment which is not within the ca­pac­i­ties of a sin­gle com­pany, you need to work with the es­tab­lished sci­en­tific in­sti­tu­tions that have been do­ing it for years, and this is pre­cisely what we would like to do.

What is the rate of pen­e­tra­tion of ma­chine learn­ing in South­east Europe and how does our re­gion com­pare to the rest of the world in this de­part­ment? What re­cep­tion do these tech­nolo­gies see on the part of the busi­ness or­gan­i­sa­tions and the state in­sti­tu­tions?

The re­gion of South­east Europe (SEE), and Bulgaria in par­tic­u­lar, is quite well po­si­tioned in ma­chine learn­ing. The Bul­gar­ian Acad­emy of Sciences and the other re­gional sci­en­tific or­gan­i­sa­tions are in­vest­ing in com­pu­ta­tional lin­guis­tics, it is be­ing taught as an aca­demic sub­ject, various events are be­ing or­gan­ised and the sci­en­tific com­mu­nity it­self is in­vest­ing in this type of tech­nolo­gies.

There are also some star­tups that op­er­ate in the area of com­pu­ta­tional lin­guis­tics and ar­ti­fi­cial in­tel­li­gence - some from a more sci­en­tific per­spec­tive and other not, some us­ing ready-made so­lu­tions and oth­ers de­vel­op­ing their own ones. In SEE, the IT and out­sourc­ing sec­tors in gen­eral are very well de­vel­oped, as is the devel­op­ment of web­sites, how­ever, there is still un­tapped po­ten­tial in the area of ar­ti­fi­cial in­tel­li­gence devel­op­ment.

Re­gret­tably, a large part of these so­lu­tions are be­ing per­ceived as sci­ence fic­tion and that is why our clients are mostly western com­pa­nies, usu­ally me­dia com­pa­nies or com­pa­nies in­volved in data pro­cess­ing and in­tel­li­gence. For the time be­ing, the re­gional busi­ness, apart from the star­tups, re­lies mainly on con­ven­tional tech­nolo­gies. How­ever, as the skills and tech­nolo­gies de­velop, this is bound to change.

Au­to­mated pro­cesses are quicker and cheaper, un­block­ing com­pa­nies' cre­ative po­ten­tial. Out tech­nolo­gies in­crease read­ers’ en­gage­ment, al­low­ing me­dia to di­ver­sify their busi­ness mod­els.

Vladimir Petkov, CEO, Identrics

Identrics is an in­no­va­tive tech­nol­ogy provider of au­to­mated jour­nal­ism, data min­ing and se­man­tic web so­lu­tions, whose ded­i­cated team builds and trains tech­nol­ogy mod­els based on a client’s own data set and fit­ting his needs.

Newspapers in English

Newspapers from Bulgaria

© PressReader. All rights reserved.