Heroconf London 2018_Automating Search Query Processing
1 like835 views
The document discusses automating search query processing from simple tactics to machine learning approaches. It outlines understanding challenges, getting started with simple tools, and making an informed decision about automation. It then details steps to implement entity recognition, starting simply and expanding use of Google APIs, Python libraries, and supervised machine learning to improve accuracy beyond 80% for processed queries. The overall goal is to help automate more time-consuming PPC tasks and scale query expansion and automation.
![{result: {
'First Name': 'Philipp',
'Last Name': 'Mainka',
'@type: [
'Person',
'Google Specialist'
],
Employer: 'Sixt',
Position: 'Head of SEA',
Contact:
'linkedin.com/in/philipp-
mainka-12b7b543/'}
THANK YOU - Your Nerd Questions Please.
{result: {
'First Name': 'Christopher',
'Last Name': 'Gutknecht',
@type: [
'Person',
'Google Specialist'
],
Employer: 'norisk',
Position: 'Head of OM',
Contact:
'linkedin.com/in/chrisgutknecht/'}](https://image.slidesharecdn.com/heroconflondon2018deckpmcgv4compressed-181022111019/85/Heroconf-London-2018_Automating-Search-Query-Processing-34-320.jpg)
Heroconf London 2018_Automating Search Query Processing
- 1. Automating Search Query Processing From Simple Tactics to Machine Learning Approaches Philipp Mainka & Christopher Gutknecht
- 2. Philipp Mainka Head of SEA @ Sixt 9 years in PPC 13+ Awarded Team Focus on Travel Munich-based Performance Marketing Data Nerd Christopher Gutknecht Head of Online Marketing @ norisk 10 years in PPC SEMY Award Jury Member Focus Ecommerce & Retail Munich-based Dad of 2,5 yr old Self Taught Rookie-Dev Two Nerds - Same Problem
- 3. Your Learning Outcome of This Talk 1 2 3 Understanding the challenges of query automation How to get started with little helper tools Being able to make an informed make-or-buy decision
- 4. Current Trend: Managed by Google vs Control? Automated Extension Smart Bidding Responsive Search Ads Dynamic Search Ads Suggested Ads Variant 'Exact' Universal App Campaigns ...
- 5. It’s Not About This Happening… YOU
- 6. It’s About Your Role in the Future of PPC HEAD CHEF WAITEROR
- 7. Time Campaign Mgt Bid Mgt Reporting & BI The Most Time-Consuming Tasks in PPC
- 8. Source: youtu.be/TcYZi9rEvgo The Search Query Workflow - Simple, Right?
- 9. The Search Query Reality: Entities Everywhere! COLOR GENDER PLUS SIZE SIZES CATEGORY CUSTOM MATERIAL Entities define the place in the account structure
- 10. Our Two Major Battlefields of PPC RETAIL + More Query sources + More Internal Structure + More Shopping & DSA TRAVEL + More Entity Combinations + More Entity Hierarchies + External APIs needed (Geo)
- 11. Matchtpye s New Keywords Entities recognition Bring into structure Find URL Automatic New Query Processing: Step-by-Step Write AdsCalculate Bids Matchtype & Settings
- 12. Use your entities Ask Google Starting with ML Supervised ML Entity Recognition: Start Simple, then Expand
- 13. ● Text normalization: lowercase, ascii, word sort, add nospace ● Partial matches: Partial, fuzzy, wordstem, typo, synonym ● False matches: withinString ● Cross-entity match: 'dress' = title or category? Product Feed Entities String Comparison Benchmark 80%
- 14. Entity Recognition in Retail: Step-by-Step FullMatch PartialMatch ● MaxMatch ○ Match 1 ○ Match 2 1. Brand? stripFromQuery 'jersey stretch plus size tops' 2. Category? ... n. Custom plusSize: plus size, oversize discount: sale, clearance year: 2018, 2019 … : ResultObject isTrue FuzzyMatch Example Script: bit.ly/hero_sqa
- 15. Entity Detection: Example Output (Video)
- 16. Use your entities Ask Google Starting with ML Supervised ML Start Simple and Expand: Ask Google
- 17. cloud.google.com/natural-language/ #AskGoogle 1: Try Google's Entity Recognition Query Cloud Natural Language API Get Entities
- 18. ● Example Call: here #AskGoogle 2: Validate Geo Entities URL Fetch Parse Results Validate Geo Entities ● Example Script: https://goo.gl/FLdtK4
- 19. ● Example Call: bit.ly/hero_sqa ● Example Script: bit.ly/hero_sqa Call Suggest API Get Edit Distance #AskGoogle 3: Typo Recognition
- 20. ● Example Script: bit.ly/hero_sqa #AskGoogle 4: Validate Synonyms Query Scrape Related Searches Compute Intersection
- 21. site: Query Scrape URLs ● If Onsite-Search not an option ● Free Tool: oneproseo.com/landingpagefinder/ (Limit: 100 req/day) ● Example script with Scraping API: bit.ly/hero_sqa #AskGoogle 5: Find a Target URL Semantic Validation
- 22. Use your entities Ask Google Starting with ML Supervised ML Start Simple and Expand: Enter Machine Learning
- 23. Why Python Is Amazing: Example
- 24. fuzzyWuzzy String Similarity Language Model Context Modeling Python NLP Packages to go Beyond 80% Accuracy
- 25. How Python Can Be Triggered From Ads Scripts SCRIPTS CLOUD FUNCTIONS ... ... ● Example Script: bit.ly/hero_sqa
- 26. Use your entities Ask Google Starting with ML Supervised ML Start Simple and Expand: Supervised ML
- 27. Rule based analysis: rent a van in miami Why Context Matters: Rules often don’t work VEHICLE_TYPE: van LOCATION: van
- 28. With machine Learning: rent a van in miami Why Context Matters: Resolve Ambiguity VEHICLE_TYPE: van LOCATION: van
- 29. Handling Ambiguous Entities Group into Sub-Entities Prioritisation & Deduplication
- 30. Clean Training Data I think it is a VEHICLE_TYPE bristol car & van hire Data Validation - The App Way!
- 31. Improving Model Accuracy Beyond 90% Accuracy for 50k processed queries80% 85% 90% + Stemming + Lemmatization + Variants + Ambiguity + Supervised + Language Model
- 32. How to Get Started with Query Automation Be the Head Chef: Start Experimenting!1 2 3 Start Simple and Expand Customise and Scale
- 33. Query Expansion Entity Exploration Value Prediction The Road Ahead: Automating Search Queries
- 34. {result: { 'First Name': 'Philipp', 'Last Name': 'Mainka', '@type: [ 'Person', 'Google Specialist' ], Employer: 'Sixt', Position: 'Head of SEA', Contact: 'linkedin.com/in/philipp- mainka-12b7b543/'} THANK YOU - Your Nerd Questions Please. {result: { 'First Name': 'Christopher', 'Last Name': 'Gutknecht', @type: [ 'Person', 'Google Specialist' ], Employer: 'norisk', Position: 'Head of OM', Contact: 'linkedin.com/in/chrisgutknecht/'}