1 / 34

R For A Data Driven SEO Workflow

R For A Data Driven SEO Workflow. Sam Collins – SEO Manager Sam Hall – Data Scientist. What does TravelSupermarket do?. “To reduce the stress of booking travel by providing the best UK comparison experience”. Today we are going to take you through…. How to solve a problem like SEO

soniaj
Download Presentation

R For A Data Driven SEO Workflow

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. R For A Data Driven SEO Workflow Sam Collins – SEO Manager Sam Hall – Data Scientist

  2. What does TravelSupermarket do? “To reduce the stress of booking travel by providing the best UK comparison experience”

  3. Today we are going to take you through… • How to solve a problem like SEO • Turning noise into insight • Proving value in what we’ve learnt

  4. What’s our problem?

  5. Why is Search Engine Optimisation Important To Us? ~60% ~50% Revenue Visitors

  6. Do more SEO

  7. The problem is, SEO is difficult

  8. Firstly, it has a lot of moving parts User Signals Keywords Trustworthiness Semantic Relevance 200+ Site Speed UX Backlinks Brand Salience Reviews

  9. Secondly, your website doesn’t operate in a vacuum

  10. Thirdly, travel is a big marketplace 60,000 keywords

  11. But you work at it…

  12. But you work at it…

  13. …and then Google changes the rules… User Signals Keywords Trustworthiness Semantic Relevance Site Speed MEDIC Site Profiling Authorship UX Backlinks Brand Salience Reviews

  14. ….and doesn’t say how Cool, thanks Google

  15. Which leaves you with one question Page B Google Update WHY?!! Performance Page A Time

  16. Here’s where R comes to the rescue

  17. R to the rescue! What do we know? Ranking Factors Winners & Losers Date of Change So, we can build a model

  18. Cutting through the noise.

  19. I R Scientist Ranking data Build models Watch out Google!

  20. Reframing as a supervised learning problem we can understand what features are associated with pages that were affected by the update. Third Party SEO tools urlprofiler Google Search Console API Google Analytics Web scraping Chrome CLI CRAN: SearchConsoleR, googleAnalyticsR

  21. We can explain the output of our model with the help of SHAP Negative impact on model prediction Positive impact on model prediction SHAP (SHapelyAdditive exPlanations) values help explain each features contribution to the prediction CRAN: SHAPforxgboost https://github.com/pablo14/shap-values

  22. Higher session duration is good Landing pages with a session durations > 5 minutes had negative impact on model prediction – less likely to see decrease

  23. How do we keep people on our site? Old destination template New destination template How do we test SEO changes?

  24. Does it work?

  25. We can’t use randomised control trials… Champion (current) 50% = split Challenger (new) 50%

  26. But we can model the counterfactual to estimate treatment effect SEO Clicks (proxy for ranks) Treatment actual All holiday pages Impact! Treatment predicted (CONTROL) Split into representative page groups Use control to predict treatment Control actual Treatment Control Change made Time Make the change to the treatment pages CRAN: CausalImpact, bsts

  27. What makes this method perfect for SEO at TravelSupermarket? • Accounts for seasonality • Valid when Google comes along and messes things up • Like other channels, success can be measured with a monetary value

  28. Here’s one we made earlier… Hypothesis: Adding price to the meta title of the page will improve position & click through rate.

  29. It worked. Treatment Control cumulative uplift

  30. So, maybe SEO isn’t that difficult…

  31. In Summary: We used R to… SEO is hard Cutting through the noise ££££

  32. Done anything similar? Come say hi!

  33. References: • S.M Lundberg, Su-In-Lee (2017). A Unified Approach to Interpreting Model Predictions. http://papers.nips.cc/paper/7062-a-unified-approach-to-interpreting-model-predictions.pdf • K.H. Brodersen, F. Gallusser, J. Koehler, N. Remy, S. L. Scott, (2015). Inferring Causal Impact Using Bayesian Structural Time-Series Models. https://research.google.com/pubs/pub41854.html. • https://www.distilled.net/resources/what-is-seo-split-testing/ • https://cran.r-project.org/web/packages/searchConsoleR/searchConsoleR.pdf

More Related