Good bye conversion rate - a smarter way to optimising Search Engine Result Pages
-
Upload
martin-gaedke -
Category
Data & Analytics
-
view
14 -
download
0
Transcript of Good bye conversion rate - a smarter way to optimising Search Engine Result Pages
Faculty of Computer Science Professorship Distributed and self-organizing Systems
Good bye conversion ratea smarter way to optimising Search Engine Results Pages
MartinGaedke.comVSR.Informatik.TU-Chemnitz.de
///// IS-SWIS 2017, Rotterdam, February 9, 2017///
(simplified) Motivation
4
Evaluation Method Efficient? Effective?
User Testing ✖ ✔
Usability Inspection ✖ ✔
Split Testing ✔ ✖
MartinGaedke- http://vsr.informatik.tu-chemnitz.de
Split Testing
5
Buy this Buy this
MartinGaedke- http://vsr.informatik.tu-chemnitz.de
VersionA VersionB
Conversion Rate:proportion of visitors to a website who take action due to subtle requests from marketers, SEOs, (semantic) data providers etc.
The problem with Split Testing
6MartinGaedke- http://vsr.informatik.tu-chemnitz.de
(source:Wikipedia.com)
8
$$$ ≠ Usability
Jakob Nielsen: Putting A/B Testing in Its Place, http://www.nngroup.com/articles/putting-ab-testing-in-its-place/
MartinGaedke- http://vsr.informatik.tu-chemnitz.de
Idea: Rethink the target measure from the users point of view (i.e. development & user – not sales) : Usability as a measure for split testing
11
Idea: Infer quantitative measure ofusability from user interactions.
MartinGaedke- http://vsr.informatik.tu-chemnitz.de
Usability-based Split Testing
12
Interface A Interface B
Questionnaire
[WaPPU – UsabilitybasedA/B-TestingSpeicher,Both,Gaedke,ICWE2014]
MartinGaedke- http://vsr.informatik.tu-chemnitz.de
WaPPUUsabilitySystem
(NaiveBayes classier for thelearning)
Modelfor
Usability
Usability-based Split Testing
13
Interface A Interface B
[WaPPU – UsabilitybasedA/B-TestingSpeicher,Both,Gaedke,ICWE2014]
MartinGaedke- http://vsr.informatik.tu-chemnitz.de
WaPPUUsabilitySystem
TRAINEDModel
forUsability
Usability Model allows for:f(cause) = score
(cause is part of usability)
14MartinGaedke- http://vsr.informatik.tu-chemnitz.de
Sure!
f(cause) = score this function is not bijective
(Not possible to infer optimizations from usability scores )
Limitations?
16MartinGaedke- http://vsr.informatik.tu-chemnitz.de
19
SERP Optimization Suite (S.O.S.)
WaPPUfor evaluation
Catalogof best practices
MartinGaedke- http://vsr.informatik.tu-chemnitz.de
[S.O.S.:DoesYourSearchEngineResultsPage(SERP)NeedHelp?Speicher,Both,Gaedke,CHI 2015]
20
fu (score) = {C, C'}
u = usability item numberC = potential causesC' = countermeasures
MartinGaedke- http://vsr.informatik.tu-chemnitz.de
S.O.S. example: "We have a problem with Distraction"
21MartinGaedke- http://vsr.informatik.tu-chemnitz.de
28
“Find a birthday present for a goodfriend that does not cost more
than 50 Euros.”
MartinGaedke- http://vsr.informatik.tu-chemnitz.de
32
Resultsold SERP new SERP
informativeness -0.17 â < -0.02 â
understandability 0.34 â < 0.45 â
confusion 0.30 â < 0.38 â
distraction* 0.36 â < 0.62 ä
readability 0.45 â < 0.52 â
information density*
0.04 â < 0.43 ä
accessibility 0.06 â < 0.07 â
overall usability* 59.86% â < 67.50% ä
MartinGaedke- http://vsr.informatik.tu-chemnitz.de
n S.O.S. significantly improved SERP usabilityn Approach developed using HCD/DT
n obtaining usability scores through WaPPU andn recommendations for optimization from a
catalog for suboptimal scoresn http://vsr.informatik.tu-chemnitz.de/demo/SOS
Conclusions
34MartinGaedke- http://vsr.informatik.tu-chemnitz.de
n Catalogs for other categories of Web pagesn Creating trained model for web page and site
categoriesn Crating more generic and reusable catalogsn Reducing the learning time
n Applying the approach to the quality of linked enterprise data
n Improving: incorrect, incomplete etc. datan Dealing with: Co-evolution, Coherence etc.n Data Quality à Fit for business usen LEDS project:
http://www.leds-projekt.de/
Future Work
35MartinGaedke- http://vsr.informatik.tu-chemnitz.de
36MartinGaedke- http://vsr.informatik.tu-chemnitz.de
Data Quality canbeinterpretedasthedegreetowhichdatafitsto current requirements
𝐷𝑄#$%&'(& = ∑ ∑ 𝜔-'. / 𝑓𝑢𝑙𝑓𝑖𝑙𝑙 𝑟𝑒𝑞, 𝑜𝑏𝑗�-'.<
�='>'?< +𝜑
37MartinGaedke- http://vsr.informatik.tu-chemnitz.de
FirstInspiringResultsOutputofFitnessresultsofourQualityAssessmenttoolwiththemeansofthedataqualityvocabulary(dqv)
Feelfreetovisit:www.leds-projekt.de
Chemnitz University of Technology #VSR
Thank You!
VSR.Informatik.TU-Chemnitz.de
@gaedke /gaedke
38MartinGaedke- http://vsr.informatik.tu-chemnitz.de