Speculation Rules APIを用いたページ・広告表示の高速化によるメディア・広告KPIの改善への取り組み / Efforts to Improve Media and Advertising KPIs by Speeding Up Page and Ad Display Using Speculation Rules API

&GGPSUTUP*NQSPWF.FEJBBOE"EWFSUJTJOH ,1*TCZ4QFFEJOH6Q1BHFBOE"E%JTQMBZ 6TJOH4QFDVMBUJPO3VMFT"1* -:$PSQPSBUJPO *NBNJZB.BTBLJ ,JSBLV5PNPLJ

Imamiya Masaki Kiraku Tomoki Yahoo! JAPAN News Engineer Yahoo! JAPAN
Ads Engineer

*OUSPEVDUJPO

8IZ8F*NQSPWF8FC1FSGPSNBODF

#FUUFSXFCQFSGPSNBODF NBLFTFWFSZPOFIBQQZ

#FUUFSXFCQFSGPSNBODF NBLFTFWFSZPOFIBQQZ User Ad Publisher Contents Provider Us

IUUQTUFDICMPHZBIPPDPKQFOUSZ $BTF4UVEJFTPG:BIPP+"1"//FXT $PSF8FC7JUBMT IUUQTUFDICMPHZBIPPDPKQFOUSZ #'$BDIF 174FTTJPO "E3FWFOVF

1SFGFUDI1SFSFOEFS

1SFGFUDIJOHSFGFSTUPUIFQSBDUJDFPG TQFDVMBUJWFMZGFUDIJOHEPDVNFOUTPS TVCSFTPVSDFTJOUIFCBDLHSPVOEGPS QBHFTUIBUUIFVTFSJT MJLFMZ UPOBWJHBUF UPJOUIFOFBSGVUVSF 1SFGFUDI1SFSFOEFS 1SFGFUDI 1SFSFOEFS
IUUQTEFWFMPQFSNP[JMMBPSHFO 64EPDT(MPTTBSZ1SFGFUDI 1SFSFOEFSJOHSFGFSTUPUIFQSBDUJDFPG TQFDVMBUJWFMZQSFGFUDIJOHBOESFOEFSJOH QBHFTUIBUUIFVTFSJTMJLFMZUPOBWJHBUF UPJOUIFOFBSGVUVSF UIFCSPXTFS SFOEFSTUIFQBHFJOUIFCBDLHSPVOEJO XIBUJTFGGFDUJWFMZBOJOWJTJCMFTFQBSBUF UBC 1SFSFOEFSJOHJODMVEFTEPXOMPBEJOH BEPDVNFOUTTVCSFTPVSDFTBOESVOOJOH BTTPDJBUFE+BWB4DSJQU IUUQTEFWFMPQFSNP[JMMBPSHFO 64EPDT(MPTTBSZ1SFSFOEFS

1SFGFUDIWT1SFSFOEFS 5SBOTJUJPOGSPN1BHF"UP1BHF# 1BHF" 1BHF# Without Prefetch/Prerender Prefetch Prerender

1SFGFUDIWT1SFSFOEFS 5SBOTJUJPOGSPN1BHF"UP1BHF# 1BHF" 1BHF# Fetch Page B Render Page B
View Page B Without Prefetch/Prerender Prefetch Prerender

1SFGFUDIWT1SFSFOEFS 5SBOTJUJPOGSPN1BHF"UP1BHF# 1BHF" 1BHF# Prefetch Page B Render Page B
View Page B Fetch Page B Render Page B View Page B Without Prefetch/Prerender Prefetch Prerender

1SFGFUDIWT1SFSFOEFS 5SBOTJUJPOGSPN1BHF"UP1BHF# 1BHF" 1BHF# Prefetch Page B Prefetch Page B
Prerender Page B View Page B Render Page B View Page B Fetch Page B Render Page B View Page B Without Prefetch/Prerender Prefetch Prerender

4QFDVMBUJPO3VMFT"1* <script type="speculationrules"> { "prerender": [{ "where": { "and": [{
"href_matches": "/articles/*" }, { "not": { "selector_matches": ".no-prerender" } }] }, "eagerness": "moderate" } ] } </script>

"href_matches": "/articles/*" }, { "not": { "selector_matches": ".no-prerender" } }] }, "eagerness": "moderate" } ] } </script> 1SFSFOEFSUBSHFU

"href_matches": "/articles/*" }, { "not": { "selector_matches": ".no-prerender" } }] }, "eagerness": "moderate" } ] } </script> 1SFSFOEFSUJNJOH

4QFDVMBUJPO3VMFT"1* 1SFSFOEFSUJNJOH JNNFEJBUF "TTPPOBTQPTTJCMF FBHFS 5IJTDVSSFOUMZCFIBWFTJEFOUJDBMMZUP UIF JNNFEJBUF TFUUJOH NPEFSBUF
)PWFSPWFSBMJOLGPSNJMMJTFDPOET DPOTFSWBUJWF 1PJOUFSEPXO &BHFSOFTTPQUJPO

,FZ1PJOUTPG4QFDVMBUJPO3VMFT"1* 5VOFQSFSFOEFSUBSHFUBOEQSFSFOEFSUJNJOHUPSFEVDFVOOFDFTTBSZSFOEFS 1BHF" 1BHF# 1BHF$ 1BHF% 1SFSFOEFSBOEUSBOTJUJPO 1SFSFOEFSCVUOPUSBOTJUJPO 1SFSFOEFS BOE
USBOTJUJPO 1SFSFOEFS 1SFSFOEFSIJUSBUF

&WBMVBUJOH1SFSFOEFS*NQBDU UISPVHI"#5FTUJOH

0WFSWJFXPG"#5FTUJOH 1SFSFOEFSGSPN:BIPP+"1"/5PQ1BHFUP/FXT 1SFSFOEFS FBHFSOFTTNPEFSBUF NTIPWFS

8IZEFTLUPQWFSTJPO IUUQTDBOJVTFDPNNEOIUNM@FMFNFOUT@TDSJQU@UZQF@TQFDVMBUJPOSVMFT@QSFSFOEFS • PG:BIPP+"1"/VTFSTVTFCSPXTFSTUIBUTVQQPSUUIF 4QFDVMBUJPO3VMFT"1* • 8BOUFEUP1SFSFOEFSUIFQPJOUFSPQFSBUJPOBTBUSJHHFS

5FTUUIFFGGFDUPGQSFSFOEFS QBHFUSBOTJUJPOUJNF 8JUI1SFSFOEFSTFD 8JUIPVU1SFSFOEFSTFD ˞5FTUFEXJUI$ISPNF%FW5PPMTXJUIYTMPXEPXOUISPUUMJOHPO$16

3FTVMUTPG"#5FTUJOH .FEJBO1BHF-PBE 5JNF TFD 6TFS#FIBWJPS.FUSJDT 17 4FTTJPO%VSBUJPO /P EJGGFSFODF
"E3FWFOVF /P EJGGFSFODF

3FTVMUTPG"#5FTUJOH 5IFQSFSFOEFSIJUSBUFXBT 1BHF" 1BHF# 1BHF$ 1BHF% 1SFSFOEFSBOEUSBOTJUJPO 1SFSFOEFSCVUOPUSBOTJUJPO ʜ

%JTDVTTJPO 'BTUFS CVUOPJNQBDUPO6TFS#FIBWJPS.FUSJDTPS"E3FWFOVF *NQSPWFNFOUXBTNBSHJOBMCFDBVTFJUXBTBMSFBEZGBTU /PTBMFTJNQBDUEVFUPJOBCJMJUZUPQSFSFOEFSBET /FYU"DUJPO "ETBMTP1SFSFOEFS

%JTDVTTJPO 1SFSFOEFSIJUSBUFXBTBTMPXBT 5IFTQBDFCFUXFFOUIFOFXTIFBEMJOFTJTUPPOBSSPX SFTVMUJOHJOB MPXFSIJUSBUF /FYU"DUJPO 5FTUPOBOPUIFSQBHF • 4JOHMFMJOLMPDBUJPO •
.PTUMJLFMZQBHFUSBOTJUJPOJTLOPXOJOBEWBODF

Journey of Ad Prerender Second prerender A/B test

*OUSPEVDUJPOUP1BHFT6TJOH1SFSFOEFS above AD below AD

above AD below AD *OUSPEVDUJPOUP1BHFT6TJOH 1SFSFOEFS

-JOL -JOL DMJDL USBOTJUJPO "EMJCSBSZ MPBE "E3FRVFTU )BOEMJOH "E%JTQMBZ 1SPDFTT
1BHFMPBE "% 3FOEFSJOH'MPXPG"%

3FOEFSJOH'MPXPG"% -JOL -JOL DMJDL USBOTJUJPO "EMJCSBSZ MPBE "E3FRVFTU )BOEMJOH "E%JTQMBZ
1SPDFTT 1BHFMPBE "%

3FOEFSJOH'MPXPG"%

above AD below AD

CFMPX"% below AD LoadTime: Fast AdType: Image Ad Placement: below
Position Ad Size: Medium

BCPWF"% above AD① below AD above AD② below AD

BCPWF"%① above AD① LoadTime: Slow AdType: Image Ad Placement: above
Position Ad Size: Large or Medium

BCPWF"%② LoadTime: Slow AdType: Image or Video Ad Placement: above
Position Ad Size: Large or Medium above ADᶄ

-PBE 5JNF AdType 1MBDFNFOU Size #FMPX"E 'BTU *NBHF CFMPX .FEJVN
"CPWF"E 4MPX *NBHF BCPWF -BSHFPS .FEJVN "CPWF"E 4MPX *NBHFPS 7JEFP above -BSHFPS .FEJVN

3FOEFSJOH'MPXPG1SFSFOEFS"% Page Load Ad Library Load AD Request Handling AD
Display Process above AD 800ms ~ 1500ms 500ms ~ 800ms Page Load Ad Library Load AD Request Handling AD Display Process below AD

3FOEFSJOH'MPXPG1SFSFOEFS"%

3FOEFSJOH'MPXPG1SFSFOEFS"% Page Load Ad Library Load Partial AD Request Handling
above AD Page Load Ad Library Load AD Request Handling AD Display Process below AD Remaining Processing Outside of Prerendering AD Display Process 50ms ~ 200ms 500ms ~ 800ms 600ms ~ 1450ms Not Prerender Prerender

5FTU1BUUFSO Moderate pattern Immediate pattern

5FTU1BUUFSOrNPEFSBUF above AD below AD

5FTU1BUUFSOrNPEFSBUF Link Link click transition Page load AD Link hover
Ad library load Ad Request Handling Ad Display Process Page Load

5FTU1BUUFSOr*NNFEJBUF above AD below AD

5FTU1BUUFSOr*NNFEJBUF Link Link click transition Page load AD Ad library
load Ad Request Handling Ad Display Process Page Load

3FTVMU

Improvement of Ad Display Speed above AD Normal: ~4000ms Moderate
pattern: ~1600ms Immediate pattern: ~1600ms below AD Normal: ~3000ms Moderate pattern: ~500ms Immediate pattern: ~300ms

)JHIFS)JUSBUFUIBOUIF1SFWJPVT5FTU

/P4JHOJGJDBOU$IBOHFJO$53GPS CFMPX"% below AD ※ CTR( Click Through Rate )
= click / impression * 100)

Statistically significant increase in CTR for above AD② Immediate pattern
2-3% relative increase in CTR above AD②

*OTJHIUT

Ad rendering speed doesn't directly translate to ad performance metrics.
above AD: 50–200ms faster below AD: 500–800ms faster But below AD didn’t show a significant change.

*OUSPEVDJOHQSFSFOEFSJOHGPSWJEFPBET NBZMFBEUPBOJODSFBTFJO$53 -PBE 5JNF AdType 1MBDFNFOU Size CTR #FMPX"E 'BTU
*NBHF below .FEJVN No Significant Change "CPWF"E 4MPX *NBHF above -BSHFPS .FEJVN No Significant Change "CPWF"E 4MPX *NBHFPS 7JEFP above -BSHFPS .FEJVN +2-3%

4VNNBSZ • +3% relative increase in CTR • Prerender seems
to be effective when video ads are delivered • Not effective for all ads

Speculation Rules APIを用いたページ・広告表示の高速化によるメディア・広告...

Speculation Rules APIを用いたページ・広告表示の高速化によるメディア・広告KPIの改善への取り組み / Efforts to Improve Media and Advertising KPIs by Speeding Up Page and Ad Display Using Speculation Rules API

Transcript

More Decks by LINEヤフーTech (LY Corporation Tech)

Other Decks in Technology

Featured