![@datemeT #BrightonSEO
The Gym Client/Employer Scenario
Lose weight
Enhance mood
ASSUMED USER INTENT Example Query BIGRAM HEAD TERM
I want to join a fitness
[MASK] to lose weight
and be happy
NGRAM 1: Fitness
NGRAM 2: [Mask]](https://image.slidesharecdn.com/intentgraph-deepanalysisintosearchbehaviour-201220120919/85/Intent-graph-Deep-analysis-into-search-behaviour-48-320.jpg)
![@datemeT #BrightonSEO
Predicting the [Masked] word with BERT](https://image.slidesharecdn.com/intentgraph-deepanalysisintosearchbehaviour-201220120919/85/Intent-graph-Deep-analysis-into-search-behaviour-49-320.jpg)
![@datemeT #BrightonSEO
Predicting the [Masked] word with RoBERTa](https://image.slidesharecdn.com/intentgraph-deepanalysisintosearchbehaviour-201220120919/85/Intent-graph-Deep-analysis-into-search-behaviour-50-320.jpg)
![@datemeT #BrightonSEO
Predicting the [Masked] word with XLNET](https://image.slidesharecdn.com/intentgraph-deepanalysisintosearchbehaviour-201220120919/85/Intent-graph-Deep-analysis-into-search-behaviour-51-320.jpg)
Intent graph: Deep analysis into search behaviour
- 1. Intent Graph: Deep Analysis into Search Behaviour Dateme Tubotamuno // SemanticGeek SLIDESHARE.NET/DATEMETUBOTAMUNO @DatemeT
- 3. @datemeT #BrightonSEO User Intent is usually Oversimplified
- 4. @datemeT #BrightonSEO Intents from the Eyes of Our Agenda
- 5. @datemeT #BrightonSEO Seldom from that of the user
- 6. @datemeT #BrightonSEO User Intent is in the WHY and not the WHAT
- 7. @datemeT #BrightonSEO Intent should start with the WHY Occasionally Fantasize of the ‘HOW’ e.g TRANSACTIONAL QUERIES Often Romanticise about the ‘WHAT’ Frequently minimise the ‘WHY’ e.g BERT RANKBRAIN e.g MOTIVATIONS REASONING
- 9. @datemeT #BrightonSEO INTENT GRAPH REASONING MOTIVATION
- 13. @datemeT #BrightonSEO : Reasoning is at the 3rd LEVEL of cognition
- 16. @datemeT #BrightonSEO REASONING LOWER ORDER INTENT INTUITION HIGHER ORDER INTENT
- 17. @datemeT #BrightonSEO When intent goes relational
- 18. @datemeT #BrightonSEO Deepmind joins the relational reasoning party https://deepmind.com/blog/article/neural-approach-relational-reasoning
- 19. @datemeT #BrightonSEO Comparing relative merits between two fruits https://deepmind.com/blog/article/neural-approach-relational-reasoning KIWI FRUIT MANGO
- 20. @datemeT #BrightonSEO 1 2 3 4 Current Intent Analysis = Lower Order (WHAT)
- 21. @datemeT #BrightonSEO Micro Moments = Lower Order Intent (WHAT)
- 22. @datemeT #BrightonSEO Understanding intent is key in search
- 23. @datemeT #BrightonSEO Relational Reasoning aids higher-order intent
- 24. @datemeT #BrightonSEO The deep intent behind our search Is linked to our experiences & stories
- 25. @datemeT #BrightonSEO Understanding intent should be contextual
- 26. @datemeT #BrightonSEO A BLANKET APPROACH IS COZY BUT SHALLOW
- 27. @datemeT #BrightonSEO Understanding intent is harder than you think
- 28. @datemeT #BrightonSEO My wife usually utters: “Honey the bin” What’s her intent I ask myself? Take the bin down? wet + dry waste? The bin stinks? Put some bin liner? Bin area is untidy? Brush up the area?
- 29. @datemeT #BrightonSEO Google misunderstood my intent
- 30. @datemeT #BrightonSEO Relational reasoning aids decision science Drawing up links from our experiences
- 31. @datemeT #BrightonSEO Progression to relational reasoning Percepts Relational Thinking (Lower Order) Relational Reasoning (Higher Order) Fleeting Locus - World Sensations In the moment Sensed Effortless Locus - Mind Overtime Examined Effortful Extrinsic Intrinsic https://www.nature.com/article s/npjscilearn20164/tables/1
- 32. @datemeT #BrightonSEO Reasoning is a pivotal part of the intent graph
- 34. @datemeT #BrightonSEO What are the motivations behind the query?
- 35. @datemeT #BrightonSEO Can you visualise the possible motivations?
- 36. @datemeT #BrightonSEO Maslow’s Theory of Human Motivation Source: https://www.simplypsychology.org/maslow.html
- 37. @datemeT #BrightonSEO ERG Theory condenses Maslow's 5 Needs Source:https://expertprogrammanagement.com/2018/10/erg-theory-of-motivation/
- 39. @datemeT #BrightonSEO Source: Vered Schwartz - Commonsense Reasoning for NLP: https://www.youtube.com/watch?v=InIffoMnV7k KNOWLEDGE OF ‘WHAT’ KNOWLEDGE OF ‘WHY’ & ‘HOW’ Taxonomic Knowledge (A is A B) Hierarchical Relationships Reasoning - Causes and Effects Semantic Relationships
- 40. @datemeT #BrightonSEO THE INTENT LAB: EXPERIMENT TIME
- 41. @datemeT #BrightonSEO https://github.com/EricFillion/happy-transformer TOOLS ● Python ● Happy Transformer (BERT) ● Google Colab ● Github ● NetworkX ● MatplotLib ● Scikit-Learn ● Pandas ● Numpy ● Pytorch ● Pytorch Geometric
- 42. @datemeT #BrightonSEO YOU’VE JUST WON A NEW GYM CLIENT OR EMPLOYED BY ONE
- 43. @datemeT #BrightonSEO TIME TO BUILD AN INTENT GRAPH FOR THIS CLIENT OR EMPLOYER. STARTING WITH THE USERS’ WHY!
- 44. @datemeT #BrightonSEO Motivations or Reasons for Joining a Gym Good for one’s health Reduces Anxiety Meet like-minded people Enhances mood Access to knowledge Good stress buster Lose weight Improve sleep Access to equipment
- 45. @datemeT #BrightonSEO Intents could be masked adjacent to a query
- 46. @datemeT #BrightonSEO LIKE LANGUAGE MODELS SUCH AS BERT, ROBERTA & XLNET MASKING TOKENS
- 47. @datemeT #BrightonSEO USE INTENT TO PREDICT A MASKED TOKEN IN A BIGRAM KEYWORD. ENTER: BERT, ROBERTA & XLNET
- 48. @datemeT #BrightonSEO The Gym Client/Employer Scenario Lose weight Enhance mood ASSUMED USER INTENT Example Query BIGRAM HEAD TERM I want to join a fitness [MASK] to lose weight and be happy NGRAM 1: Fitness NGRAM 2: [Mask]
- 49. @datemeT #BrightonSEO Predicting the [Masked] word with BERT
- 50. @datemeT #BrightonSEO Predicting the [Masked] word with RoBERTa
- 51. @datemeT #BrightonSEO Predicting the [Masked] word with XLNET
- 52. @datemeT #BrightonSEO ● Best prediction ● Highest Softmax ● Highest search volume ● Second best prediction ● Second highest search volume ● Worst prediction ● Lowest SoftMax ● Lowest Search Volume ROBERTA BERT XLNET
- 53. @datemeT #BrightonSEO SERP Preview of the term: “Fitness Club” Focus on location-based fitness programs
- 54. @datemeT #BrightonSEO SERP Preview of the term: “Fitness Group” Focus on the social elements of fitness programs
- 55. @datemeT #BrightonSEO BUILDING A WEIGHTED INTENT GRAPH WITH DIJKSTRA'S SHORTEST PATH ALGORITHM
- 56. @datemeT #BrightonSEO BASED ON ROBERTA’s PREDICTION OF THE MASKED TOKENS
- 57. @datemeT #BrightonSEO USING THE SOFTMAX SCORES AS THE EDGE WEIGHTS
- 58. @datemeT #BrightonSEO ‘Fitness Club’ shortest path to the intent https://github.com/gunner55/graph-intent/blob/main/Fitness_Intent_Graph.ipynb
- 59. @datemeT #BrightonSEO Graph Representation of the search terms https://github.com/gunner55/graph- intent/blob/main/Fitness_Intent_Graph_Dra w.ipynb
- 60. @datemeT #BrightonSEO PYTORCH GEOMETRIC (GRAPH NEURAL NETWORK)
- 61. @datemeT #BrightonSEO https://github.com/gunner55/graph- intent/blob/main/Fitness_Intent_Graph_Dra w.ipynb 1-10 masked token predictions by RoBERTa 11-20 masked token predictions by RoBERTa
- 62. @datemeT #BrightonSEO Toe dipping in Pytorch Geometric
- 63. @datemeT #BrightonSEO INTENT-DRIVEN MARKETERS START WITH THE USERS’ WHY
- 64. @datemeT #BrightonSEO INTENT IS AN INWARD EXPRESSION OF THE USER
- 65. @datemeT #BrightonSEO NOT AN OUTWARD DESPERATION OF A BUSINESS
- 66. @datemeT #BrightonSEO STARING AT SERP WILL NOT GIVE YOU A COMPLETE PICTURE OF INTENT
- 67. @datemeT #BrightonSEO GOOGLE IS WORKING HARD TO SOLVE SEARCH
- 68. @datemeT #BrightonSEO “There is actually no end in sight in terms of when this will actually be solved." Ben Gomes - SVP, Google "Search is far from a solved Problem" Pandu Nayak - Head of Search Ranking, Google https://www.youtube.com/watch?v=tFq6Q_muwG0
- 69. @datemeT #BrightonSEO LET’S DIG DEEPER AS AN INDUSTRY