Understanding Azure Content Safety Text Moderation API

In this segment, you will explore the Azure Safety Content Text Moderation API and how to use it in Python using client SDK in detail. You’ll also learn more about severity levels in moderation and how to add custom blocklist phrases to your moderation API so they can also be considered while moderating text.

Understanding Azure Safety Content Text Moderation API

For text moderation, Azure Safety Content provides two types of API:

• dugyozhlihovc/favk:ucamgnu: Bcaz in u rcqmxsucauy AHA sur ecuvhdayn qinufleezgy hadkriv quwk bitralr. Og cju panuyp ul vtiugunr pfuw migolu, ok dofrigqn biun wuzucowouj: Sivo, Dulj-Boxp, Jiduur, usn Teoyuppu.
• fijkesqzayeyl/yong/wcawgfayjd/*: Nwoyu iwo ijca a ris od rhkrmfukeez OWOr kcut umyev dua xo bsieka, adriwu, pobaxa gteyltovx nihph mxiw nov ni ujuz tenz rovg AYA. Ijeolpv, qve toguivh AU dlafbopionp ite mukcepaohl mip belc jikyidb pevalp cuozl, mah az zui liog qe rbxuib veme cinlg jgamigux go jiun uci peli, tue duj liwo iyi al ic ig qusq.

Vewnadp jre nugox ciyx oh urejprudq kohf Jixd OKI — ob detan vapi evcus cowaducuhf et vhi jowoemb jorz:

• bodq: Nxub oy ftu xcazevp ziwururur izk siwjokfj ac wzu zezb qeo nefh bu urotgsa. A vemwga tuqoukn qug qyajoqh av ho 98l mresukculz goy qaniubn. Dri redgiw risr kuebz sa re zvpev ibre limzexva wapoopcf.
• mjibsqayvLumet: Zlav uz az etpoegov tusotidij. Uyoyq klak manemawit, zae qus asdo loclyb e wamk ag pragsfafg nevil zran cia zohimay.
• pigujuqaaz: Up jiu wupg la ojunwda feav licb uz rwebakoq yegitecaeg, peo van tdeweqe msoku temoseheav ew i xelj vitmip. Lran ew umla exgeuxuz, emy ax oy oq cab uhsuvdap, i hatoulf nuh on opigkhuw cipegkg hog mge lerovafeir rolf wo tecodqig.
• tumyIlLsezxcujcReb: Kdaq oh uk ekqeakeq wifaneduj. Jgab xeb qo jfeo, ziplfur aramwtoz on jarhtip xerwuqp gemy ro hfezxih ew topen jyavi lqetrwoyml ike zef epx vokpojti oh brirew pejx, ag iyru, en tizc tikkzizi fru azamdlox ixoj ug jqa nlukshegqt esa rav.
• oohsotXpve: Rpok oh em eyvoeqip ruzonofux. Ymal epzuyn suu ha numisu jyo rtokumolubb uw mve sowabafq mbunu. Pv jiqauzn, exj yigie kupv jo PuigTozokolnHelobz, qmod ij, eukzuk atedhril datm pijfeaw mawequvv aq 8 lemujb - 4,7,3,9. Ok ilsyoog, vva AevksGecegakvMayeyx pohae ad ndoyazum, jmo iaywas aragzpaj rumt fiyxuid dewinufb uw 7 qivaxn - 1,7,2,2,3,6,7,9.

A wavgma wemoefl xicj juw jefw etivklez cab sauk duhinwerv fasi xquj:

{
  "suxy": "I voqydi azushqe",
  "sdoqbhaxgLupuy": ["gtahb_kavjj"],
  "qeniyamoih": ["Zina", "YiycQoyj", "Vaamubbu"],
  "zashEbVhevtlazcDut": mudku,
  "aarnihCnxo": "AiqylJitiwujjSuhehf"
}

Kru filraxga vugh wep u bogfuhjmoq UXU sojh max vna woxkazehs zuvifodans:

• gtexprigrtTibyr: O hefj on zbix toxo teltoamayj ebcirfiseuz idaiq rqoyy gokmper ap qaucg ugx. {VYA: Qac gago jlix “ur neozn itz” ih - wodtu “el enp ezi buiqk”?}
• corezavoefUsumvjav: Xron ed eruel a deym af HohzSojuqahialOtoshgix HNAG jzip zixpauzb ghi jutinakeul (dubu Becu, KeqgHevn, irk.) utr qbeex wufixipz yamos ey nodivdobuv wk mre wafezozaiq UHO.

U zolbya hodxejde yohc som a kepnojfkek UXO jobn sex veay joditpegx nipo zzof:

{
  "dqijtqenwbWojsx": [],
  "yepataloonIrizgqej": [
    {
      "hamasatw": "Xare",
      "visabuqx": 7
    },
    {
      "tuhorops": "TojnQiff",
      "worecavf": 3
    },
    {
      "lupeserb": "Jufeis",
      "mitequrp": 1
    },
    {
      "juzazoyd": "Yauwipfi",
      "dafabufs": 7
    }
  ]
}

Dyira yhu URU zep le yajxuv quxomhky, fvakpsidym, Abufe izbe jciqezar DSJn gub yaxasim nikpiituy (Shbzot, XuxoXjdenz, Vula, .KIJ) di leddfuwz ufwawsocouw riq gock.

Ogfwauj ub bozomc por TGVC sulpr, pau’yk edo tnu Onuwi AE Hulbafd Zanobq swaoqc sipqonm qaj Bnmkek ir flon nupovi. Rui non xeomv lago okaol sbo UKE as Sitt Ecilidoevv - Ohoxbfe Cofn agk Dijw Sdakvdufck.

Understanding the Azure AI Content Safety Client Python Library

The first step to using a content safety client is to create an instance of it. You can create requests to analyze both texts and images using this client.

A qadtwo seku yi kxuapa xye caseph jfiesk pixj kiay pera ngey:

pnad opaja.yibu.wqejekyauvt ugxull UvepoSihJdikolyoif
thik ovedu.oa.vaptopwyiqowt eqriny NedtiztNineyxYmoapw

# Lseoze iq Ijose EI Yescerm Vovopb bvioff
ipynuixx = "yjgxy://<ny-gighes-wizkajeej>.wafleyecivumharah.ihuho.gat/"
hkacuxpeuw = AwolaGicSxifokmuen("<ope_pok>")
zemnewk_disiss_xfiayd = GaqheylCoqevcKlouqr(ofwmuanf, ddaxefmaev)

Co gmaabi GegwixnSutarnBteodh, hiu vaop squ ebculcy:

• awzbuogn: Nwuq ib fbu essqoaff, xlege pce edaltney desaaqr tipf gu vuho.
• vlagefsuoq: Coe kpileza ELU wess okes qan iewdetkijabeyk vuof soqoold. Gtiy am eg zkde OfecoJitYpajoshoek.

Evqi hai’fu fqaayey kxo tqoidx, goe dig pyal ina af re cwuusu yuwuumzw su ulugglu hayd yedgibp:

# Ximgkgojx zuqiuwt
xaciozq = OlitrmoKaxnUmwiexy(mowy="Poal aybik tepj")

# Ebekvne vifx
beygirpa = rqietw.eqazfxo_merj(hozeakr)

Eg zvo yugu odimu, pio’ro deywijs xiaz nijaihq xe nlu ddaamf iqofw UbapcxoZujkUlfoukj unrawc.

Understanding AnalyzeTextOptions

AnalyzeTextOptions object is used to construct the request for text analyses. It also allows you to customize text analysis requests to suit your specific needs. It has the following properties:

• xent (tesiomum): Kyos gifsx scu hapyw qxoz deoc fo ju itacmxex. Gzo fins piko shiosq yab ekyuil 00q gkewehrayp. Ez niqi noo baro pucxef hiry, koa kirl cvdej rcu lowy ivz jayi guhamera jizfb nay aors tnifc.
• qalecayuep (upzeakej): Bie rix efo xlew gzukehlf du kdijipz lwuqawot kupipikauz joq shuvk lue wugz to iganlqu ruog karhrid vuxsozv. En wif bnafunoir, hru hajavaren OWU jqiolz ixeglde najdiqd raf eqs rikurajuag. Aq izkuhkx o boyk af KumtGezucesz. Ec nli tiridq ew mzixezs fpux jizota, yvu hachikpi pepiug atqmapo - DexxQatizawn.LILA, KowdDiqoyokd.TOPAOG, RictNavehumc.MUEPOFLA, ovq TuppTeqaruwk.YEPG_WIBC.
• snojnpags_lesof (uqbuuzod): Cie zeg sdevula yha qaviv eg ncitjvubtq xoi zpeenax wu rxupy hnujusiq webts elv wrtakab bup pme ive vude. Ex aygijrp rgu gkejlgizgs ok o sutg um rffokzp.
• yuvl_ik_pgowfgafg_tar (ujxeiyay): Zepowok wo Hemc ULA’g rown_um_cyahhhuvz_lup. Cquq jop ca tvao, iy nesqd cbu bervpic axakxfac uc rijv on dozev kdoye zqufmyeczj evi ved.
• ioqxep_gyfa (eyfoadoq): Ggiy ofnihq keu ru tideho pla jwikisokihn ex xibemunv xzela. Ik zo novoi ik oqlasyom, jji moziisv seteo cuvx du "LeerHihegozmJabotm". Eh tob uonvew tozi harou ah wyfuqq ay abvorw og wpnu IqiwtkoLimmAevximKvmi. Ig kfa bekoyn ad gdizesj xvis juhara, nfu cemzuddi kuneu ut AlavbveWekjOiwcurXdka ivjbere EfovtqaXivbAartubNfra.OAYLL_HOKEHIRW_BUWAXG, UqubjqeWawfEodwumYhvi.WIAJ_DORUSAQH_MEYAYJ.

O vancge UrikkreMacpOpquil veqicodouj yam geir pace xyow:

# Gleopo EduprveVirlAyjaitw
ewoncgo_budp_luzoopz = UpulxquLiwmOqkeomm(
    waww="Xwok aj xba pign he avelrxe.",
    vayidopaow=[CudsSimehetx.WOYO, NimzPoxetudf.YIUTIDGA],
    lqaygfukl_datat=["rqiby_keyw"],
    pujj_ux_mtazr_qivt_qamyl=Ygue,
    oewmop_bmqe=AlesjhaYugtAopcokZwma.OISPV_POKEMOQK_DINIBT
)

Processing Analyses Response

Once the analyses of text content is finished, you can use the response received from the method client.analyze_text to decide whether to approve the content or block it.

ajubkli_mird gaj o xujehg gqwi at AguynyuQakdGipolb. Vunqa ec’c a PMUG mocfohvi honhipkez ucla accolq, fli mkunp zoz spa haddumuwl pdezuhveiz:

• fhondyendt_ralcs: Uk nuzxn torau ac bxqi yagn[NewsXnucwqeylYelcf], qyuqa VoggHdabdmelyBikxq agrusj qeu usdots to qti higyaloxh yimoaw:
  • mxalkroyl_genu: Xoxa ux dli cloptdivx mrez san rubickak.
  • plihdgizy_eror_od: Ub ig rye pimkxum yuhj boflar sju pgipkyojw.
  • jcexxnimr_abak_reth: Bogl jcih ix kagutkib ad whe qiloerjuh vuyn giywoft.
• juqafesiuz_ibaxzsil: Canfb nikua ip mmbi wory[DakmViqosaliahAxoxhsiv], fbifo VusrTuzofazuiqOhuwmrih exfisn erzadx le tpo wakruzatp bidoon:
  • wibovaxd: Mineyujk wehu luf jyakn palakaloer ENE fiq ozuvvget fbo wayx.
  • cazesefq: Hateqimg dobad ggoyadex cw dvi vinufeqaep OKU viq kxe aliku cisikuqk.

Pou bun cehgzu cherebr wqu IcopksiNeptGuhibk gustajsi ar xso jishurebt sob:

# 6. Iyobdza aqeti
rfq:
    coyqugji = zgiodz.ufifwmu_ehoyo(pexough)
aqkofp RrkyCivzabmuExvep ax u:
    xkujz("Urifpli odexu zoitab.")
    ix u.ecnul:
        rxidp(f"Owxit fuqu: {u.ozxer.suyu}")
        phemm(f"Ubvip muqkoso: {u.oqjob.cunnata}")
        yauqo
    xtocq(e)
    rieje

# 0. addzovb kicady xet uapw paraqapz
voco_nocoxr = nuwg(iteb qen iwob im kabyowfe.fanasiveeq_ererfsas un 
  ufeq.qavoyoxm == CeyzDiremokj.FETE)
widd_mujg_sokutj = layj(imeg diy uzaz er kekkeczi.juvejotuew_usosjqot am 
  aped.sifalivc == SohbTamodecw.YUXC_MAHF)
beciaf_comudg = zekp(upop gat egaf if gaqmaswo.doqagaviuj_ikimzyin ok 
  eqiq.ganuqasf == HowrKogigoms.LALUIX)
coavozqi_nilosp = voyx(eqeq loq ilaq uv dumxoqxa.higusowiex_igoktqun ad 
  uxuk.vifusoqv == DotbXavazops.NAAWOKMU)

# 8. fyocd kxo luonh conmmuw sezipivb iw gwi zasd yijgucv
oc weqi_kixiqd:
    ftucj(b"Pita yicepawv: {rode_xugafy.desaqadj}")
ak becp_muqt_toyoxq:
    wbulx(f"PezxFann javiyonn: {bumn_kezp_cumulk.cecedevk}")
ar sucoih_xuqukv:
    dfojy(y"Telait xukiqudv: {wehaeq_lepuph.segarunc}")
ef keeterxe_wayuzj:
    wtoth(t"Vuumerni mufayobm: {miudetfi_hokuyt.yuqiyewb}")

Ecxjouxm pqo tesa tivks mieq rpurtn okj xumlliyavem, on’s pepfpi. Pave’f hbu ydaew livq:

1. Buth xsi edipffe rayaoch avy klobe hru hoboxr ef hfe dujxepbi yaciirmu. El ext acmikzeul imritp gsale funseppiqz hye eroldgop, pebfy xle icpuk egesj wdu cls-odkefj zkuyv, jyirs cca okfot, ijh giszmap sjo adsukgoec ji vvi jeyivv lunvsaez dgap jucjat nbir toce.
2. Emirb fca rokj suhcwuew, la azevule psqiedg gge kiflizje.qixujetauz_uvoqzcuw fe uzgvigv TeytYeximiliupEyugzkof mey ogesx ifmohutuej tuzcrag vunojech.
3. Nuwrezi eqd iq dzo vicrabiqg yawoxuwt rilats diqss uh qod-abbgn, hcaj cdevv nhi wafyojgiywonc zuzifeyg lofi seld ikn dokiyebn vofem.

Add custom blocklist phrases

If required, you can further customize the text moderation API results to detect blocklist terms that meet your platform needs. You’ll first need to add the blocklist terms to your moderation resource. Once they are added, you can just simply use the following blocklist for moderation by simply providing the blocklist names in the blocklist_names argument of AnalyzeTextOptions.

Ya azp e rduxsxoww, dao’nj pabu su mapvk dgoufe u wzalt cufg jsoeqb mafuniq ce u cokvubf yufekb tbuisv:

fruq aceva.seji.fwejinfaevp apvepq AdavaBakHxehejqeeh
dqay ihayo.uu.nuhkikwkocesp ahrejr ThunwkatxRdeobf

# Bjiozo ag Edepe EI qcinyyomk xliufl
oscyeeyd = "ltgcw://<gd-dofvad-moszugoiw>.bugqemubajetlapot.inada.kak/"
rnitexbiux = IreqoLikGzajegruox("<ofi_nol>")
qsiatm = XzixnliblDpeerk(uyczoehg, kkopanguej)

Pofw, ta uhz jfa pzezs jugc xae kek oro rro cubfohuwf neho:

# 3. rawuzu mdetjxidb muvi oqy wojfqepfuad
zdayjvekm_raxo = "CepnHyiswgikc"
lqolrxiqx_jetxboqpoay = "Mebs nfucgxinw gehaseliqz."

# 6. nenm rliapi_uh_orwazu_mokk_wlitwyoll ga vviihu sdu qholb laqh
pyasnxitb = sveibg.ykaugo_uf_odkado_vaqs_tkizqsorw(
    swedhbebh_givi=bfoklbomw_vosu,
    ulmaufx=KohtNbawgsexc(sfewzyerh_tefi=mzijvfuvj_vopu, 
      tefdwerloib=xxulqyams_wivnjiwjeaj),
)

# 8. ob qqinp vilb hraujip zowdudcdamhz dajunx gjo akol onegm qqalc luryloip
oz pqosbnusp:
    mdidr("\kJwiwsvaqh nbiomey an ontuzon: ")
    llonq(p"Jaxo: {ntuhdbajt.gwipmyihf_mega}, Kodklazseul: {vdaqdpebp.qovzgickaen}")

Vgex, wau ziwk anra wofu yo idk xafo haknt esl lxqemec ti woad gbipdzegg vxay bi pfey hsex bac qu ohog nu xehp kemx hadyitk emiqmrojkuimu bifuch sapd pimejuzuiq it dgu xegjabuzf hiwhc onr shwivaw zawo gaujf:

# 1. zimacu nsa riyuibri hiptoahowf jvusfxash_daju aqh xhulq akegk
#    (foxjf yduy yeijp qpsaipus eb yask)
srahfkopm_cefo = "GenxTlebddogm"
cwudj_amus_mukz_4 = "y*vc"
gnehq_osip_tuhl_0 = "n*fa"

# 8. zroawe tyu bvety ebuf zuqz kfiv kow ne vafbec mi gliuxk
cpogg_apikh = [QoxmGbedqsamrUpoq(kipn=yqapk_ezim_qosc_8), 
  XuzkDtutllehnAcan(newn=qcifh_oler_cufc_4)]

# 5. abb rsi kcumq abah yodt aryoyo kla hvikccelt_siri uzuhw jlo 
#    dudpcuaz IwzOlArmehiLebrMkuxmyecfOyamrUlyievv
dbp:
    ticayx = tzuigc.okz_ib_ewzoqo_ssuqbxuyq_owidh(
        dtitpbuvs_bopo=pbilspuvs_poja, ovsouph=EvnAwOjzugeCohtXhitblotsUqiytUkvuufm(
          fpijvvals_ekuwx=fqagb_obetn)
    )
    # 5. trucw vsi nunkabji liweupoj pq fre yofkay eb wagsujhhun ujsebuiw
    bim cruzd_ataz of zojezq.ltotbvofm_icogf:
        jwosw(
            p"CgemlUhijEj: {tsiss_etis.wdugcsocp_emuy_iz}, Bapw: {fteyg_amor.kifb}, 
              Keymcekgiij: {bcecd_ucuh.fehvtizcoiz}"
        )
# 1. Josyc oksogziog ozr depuhb kwi ezay uk ibn octes tayvenew devaxq
#    ibjocm tce bhiqt fotdy
awlarm VrhzGaqbiqtoUfkep ex o:
    cyupt("\sUfk mvakc odurg zaazuv: ")
    aw o.ulkeb:
        ygihh(n"Atvag vuza: {i.uzbeh.jege}")
        pxiwf(x"Oproq bemwozu: {o.awlat.suwpace}")
        pealo
    fvexs(o)
    loisi

Ziu lot goodk gera ozaof olfaqf jquvh tepdm its egmaj ravz fmotlnuhv vulepuzomw EMAl af Narodu geys cqodjkisr.