O slideshow foi denunciado.
Utilizamos seu perfil e dados de atividades no LinkedIn para personalizar e exibir anúncios mais relevantes. Altere suas preferências de anúncios quando desejar.

Ranking Factors Going Causal: Regressions, Machine Learning, and Neural Networks

145 visualizações

Publicada em

Micah Fisher-Kirshner's slides from TechSEO Boost 2019

Publicada em: Marketing
  • Seja o primeiro a comentar

Ranking Factors Going Causal: Regressions, Machine Learning, and Neural Networks

  1. 1. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost #TechSEOBoost | @CatalystSEM THANK YOU TO THIS YEAR’S SPONSORS Ranking Factors Going Causal: Regressions, Machine Learning, and Neural Networks Micah Fisher-Kirshner, Turn/River Capital
  2. 2. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Regressions, Machine Learning, and Neural Networks Ranking Factors Going Causal
  3. 3. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost The challenge is in learning the mechanics
  4. 4. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost And nitpicking…
  5. 5. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Micah Fisher-Kirshner Director of SEO & Content @ Turn/River Capital President & Founder @ BayAreaSearch.org A modern nerd
  6. 6. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Understanding the Victory Conditions
  7. 7. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost y = 𝜷x
  8. 8. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Correlation != Causation highly significant
  9. 9. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost
  10. 10. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Regressions y = 𝜷0 + 𝜷1x1 + 𝜷2x2 + 𝜷3x3 + 𝜷4x4 + ... + 𝜷5x1x2 + 𝜷6x1x3 + 𝜷7x2x3 + ... + 𝜷8x1x2x4 + ... + 𝜷nxn + 𝜺
  11. 11. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost y = Keyword’s Position y = 𝜷0 + 𝜷1(title tag length) + 𝜷2(order in title tag) + 𝜷3(page load time) + 𝜷4(total in-body internal link value) + ... + 𝜷5x1x2 + 𝜷6x1x3 + 𝜷7x2x3 + ... + 𝜷8x1x2x4 + ... + 𝜷nxn + 𝜺
  12. 12. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost y = Keyword’s Position y = 𝜷0 + 𝜷1(title tag length) + 𝜷2(order in title tag) + 𝜷3x3 + 𝜷4x4 + ... + 𝜷5(title tag length * order in title tag) + 𝜷6x1x3 + 𝜷7x2x3 + ... + 𝜷8x1x2x4 + ... + 𝜷nxn + 𝜺
  13. 13. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost y = Keyword’s Position y = 𝜷0 + 0.001x1 + 0.002x2 + -0.003x3 + 0.05x4 + ... + 0.3x1x2 + -0.03x1x3 + -0.07x2x3 + ... + 0.1x1x2x4 + ... + 𝜷nxn + 𝜺
  14. 14. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost
  15. 15. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost
  16. 16. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost 1. Solves complex issues 2. Finds hidden connections 3. Is black box in nature Neural Networks title tag length title tag order page load time
  17. 17. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Plotting the Victorious Win
  18. 18. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost
  19. 19. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Data Collection • Correct & Consistent Mobile or desktop, quality, etc. • Removals Low correlations and limited data • Time Series Weekly changes
  20. 20. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Interactions • Positive Combinations Authority, branding, etc. • Negative Combinations Panda, spam, etc.
  21. 21. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Interpretations • Reasonable Direct traffic, Google AdWords, etc. • Outliers Wikipedia, Amazon, etc.
  22. 22. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Categorizations & Quality Checks • Grouping Search volume & ranking positions • Subsets Keyword intent & heteroskedasticity
  23. 23. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Visual Outputs • Graphs Correlation goes on scatter plots • Error Ranges Uncertainty goes on whisker plots • Regression Results Standardized data formatting
  24. 24. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Sanity Checks • Data Separation Cross-validation • Peer Review Do others see the same thing • Predictive Models Has similar accuracy going forward
  25. 25. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Taking the Time & Effort for Victory
  26. 26. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Total: ~$20,000 for the year • URL Profiler (scaling): $300 • SEMrush (rankings): $5,000 • link.fish (content): $2,000 • Majestic (external links): $5,000 • Ahrefs (internal links): $5,000 Cost
  27. 27. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Why • Algorithm Updates Knowing content or link based • Prioritization Changes and how to react • Linkbait Nothing grates engineers more
  28. 28. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost y = 𝜷0 + 𝜷1x1 + 𝜷2x2 + 𝜷3x3 + 𝜷4x4 + ... + 𝜷5x1x2 + 𝜷6x1x3 + 𝜷7x2x3 + ... + 𝜷8x1x2x4 + ... + 𝜷nxn + 𝜺 What I Planned…
  29. 29. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost What I Planned… (High Level) y = 𝜷0 + 𝜮𝜷2(CONTENT) + 𝜮𝜷3(LINKS) + 𝜮𝜷4(DOMAIN) + 𝜮𝜷5(STRUCTURE) + 𝜮𝜷6(UX) + 𝜺
  30. 30. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost What I Planned… (Content) y = ... + 𝜷2a(exact phrase title tag) + 𝜷2b(title tag length) + 𝜷2c(order in title tag) + 𝜷2d(brand name used) + 𝜷2e(title tag relevance) + 𝜷2f(exact phrase h1) + 𝜷2g(exact phrase largest font) + 𝜷2h(h1 relevance) + 𝜷2i(largest font relevance) + 𝜷2j(exact phrase body copy) + 𝜷2k(BM25 score) + 𝜷2l(Flesch-Kincaid readability) + 𝜷2m(exact phrase first 100 words) + 𝜷2n(exact phrase URL) + 𝜷2o(URL relevance) + 𝜷2p(exact phrase title tag * order in title tag) + 𝜷2q(exact phrase title tag * largest font * URL) + 𝜷2r(exact phrase title tag * BM25 score) + ... + 𝜺
  31. 31. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost So… • Internal Links Checker API Ahrefs: I need you! • SEO Testing Platform Distilled ODN: Need a lot of traffic • Build-a-Google Custom search: Need engineers
  32. 32. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Enjoying the Taste of Victory
  33. 33. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Demand Better • Use the scientific fields’ toolsets • QA your data before analyzing • Be reasonable and avoid grandiose claims • Analyze where one can dive deeper • Use plots, not charts • Get the study reviewed before publishing • VICTORY!
  34. 34. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Citations & Personal Board Gaming Tips • https://scikit-learn.org/stable/tutorial/ machine_learning_map/index.html • https://e360.yale.edu/features/flooding-hot- spots-why-seas-are-rising-faster-on-the- u.s.-east-coast • https://www.scientificamerican.com/article/ graphics-that-seem-clear-can-easily-be- misread/ • https://sdm.lbl.gov/snta/2018/SNTA18_KW ON.pdf • https://www.mdpi.com/journal/futureinternet /special_issues/SEO • Axis & Allies Invade Spain with the United Kingdom • Star Trek Ascendancy Monitor Romulan research continually • Tridimensional Chess Look always at the board top-down • Suburbia Build income to max then go reputation • Galaxy Trucker Focus on weapons and engines over exposure • Chaos of the Old World Don’t let Khorne and Nurgle team up
  35. 35. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Huge thanks to Christian, our Data Scientist – Twitter, LinkedIn, and the Internet: micahfk
  36. 36. Micah Fisher-Kirshner | @micahfk | #TechSEOBoost Thanks for Viewing the Slideshare! – Watch the Recording: https://youtube.com/session-example Or Contact us today to discover how Catalyst can deliver unparalleled SEO results for your business. https://www.catalystdigital.com/

×