Machine Learning for SEOs - MozCon 2018
Download as PPTX, PDF13 likes4,120 views
1. Machine learning is a subset of artificial intelligence that allows computers to learn without being explicitly programmed by combining statistics and programming. 2. The document discusses how machine learning can be applied to search engine optimization tasks like content creation, keyword research, and traffic predictions. 3. Resources for learning machine learning are provided, including courses, books, and tools to get started with building models.
![Yearning Learning (free book preview by
Andre Ng)
Neural Networks & Deep Learning
Correlation vs Causation (by Dr. Pete!)
Exploring Word2Vec
The Zipf Mystery
BigML
Targeting Broad Queries in Search
Project Mosaic Books
How to eliminate bias in data driven
marketing
TensorFlow Dev Summit 2018 [videos]
NLP Sentiment Analysis
Talk 2 Books
Image-Net
The Shallowness of Google Translate
TF-IDF
LSI
LDA
Learn Python
Massive Open Online Courses
Coursera Machine Learning
Advanced Resources](https://image.slidesharecdn.com/mozcon2018machinelearningforseos-180712190613/85/Machine-Learning-for-SEOs-MozCon-2018-83-320.jpg)
Machine Learning for SEOs - MozCon 2018
- 1. Machine Learning for SEOs Britney Muller @BritneyMuller
- 16. Image Spark Chrome Plugin
- 18. 65% Probability this is Rand!
- 24. Isn’t ML SO Exciting!?
- 25. 1.Let’s break down Machine Learning 2. Discuss other ML applications for SEO 3. Tools & Resources
- 26. What is Machine Learning? Machine Learning is a subset of AI that combines statistics & programming to give computers the ability to “learn” without explicitly being programmed.
- 28. Lady Ada Lovelace Inspired Machine Learning 1843. Alan Turing later referred to it as “Lady Lovelace’s objection.”
- 32. Unsupervised Supervised Home size vs selling price Duck or Snake?Animal Types Three Common Models:
- 34. Linda Pumpkin
- 35. But, how do ML models get smarter?
- 37. Overfitting is a common problem in ML
- 39. If Machine Learning was a car, data would be the fuel
- 42. 1. What is Machine Learning & why you should care? 2. How can you apply ML to SEO? 3. Tools & Resources
- 45. Templates:
- 50. Recommendation Models Are Evolving
- 51. !?
- 52. !!
- 53. Pages per session went up 11.07% YoY
- 54. Help visitors stick around & get more value
- 55. Help visitors stick around & get more value
- 57. Writing Meta Descriptions Sucks
- 64. @jroakes @GraysonParks Grayson Parks Writer, programmer, constant learner. Digital marketer, husband, golden retriever owner. Words and data are my Passions. #SEO @AdaptPartners GraysonParks.com JR Oaks Hacker, Technical SEO, NC State fan, co-organizer Of Raleigh & RTP Meetups, as well as Search Engine Land author codeseo.io
- 66. 1. Assist with deploying AWS Lambda. --Several steps will affect the cost & security. 2. Extract the content of the webpage using the library Goose3 (a Python library w/BeautifulSoup). 3. Summarize the content using summa (or another summarizing library/model) 4. Create a Lambda Function. a. Package the files for AWS Lambda & install the dependencies (in this case Goose3 and summa, etc) into a folder along with what is called a handler file. The handler file is what Lambda calls to run your script. b. Here is the packaged Lambda function (including the dependencies): https://s3.amazonaws.com/ap-lambda-functions/meta_summa.zip 5. Once the zip file is deployed to AWS as a Lambda function, you should get a URL to access the API that looks like: https://XXXXXXXX.execute-api.us-east-1.amazonaws.com/v1/ap_meta_descriptions Find a developer familiar with AWS to:
- 67. function pageDescription(url, length) { if (typeof length == 'undefined' || !length || length < 1){ var endpoint = 'https://XXXXXXXX.execute-api.us-east-1.amazonaws.com/v1/ap_meta_descriptions?url=' + url; }else{ var endpoint = 'https://XXXXXXXX.execute-api.us-east-1.amazonaws.com/v1/ap_meta_descriptions?url=' + url + "&len=" + length; } var response = UrlFetchApp.fetch(endpoint); var text = response.getContentText(); var data = JSON.parse(text); if (data){ return data.meta_description } } Copy & Paste like a badass in GSheets! =pageDescription(A2, 150)
- 70. Use Text Summarization Algorithms to Help Aid the Writing of Meta Descriptions (GitHub Repo)
- 71. Finding ranking opportunities Title tag optimization Keyword opportunity gaps Client reports Finding common question opportunities Content creation Log file analysis Ranking predictions Site crawl opportunities GSC data analysis Rich customer understanding Traffic predictions Ranking factor probabilities User engagement Other SEO Opportunities with Machine Learning:
- 72. 1. What is Machine Learning & why you should care? 2. How can you apply ML to SEO? 3. Tools & Resources
- 73. 1. Collect & clean dataset 2. Build your model 3. Train 4. Evaluate 5. Predict Most of the work A few lines of code One line One line One line How to build your first ML model:
- 81. CPU > GPU > TPU
- 82. Google’s Machine Learning Crash Course Google Code Labs Colab Notebooks Learn With Google AI Image-net.org Kaggle Getting Started Resources
- 83. Yearning Learning (free book preview by Andre Ng) Neural Networks & Deep Learning Correlation vs Causation (by Dr. Pete!) Exploring Word2Vec The Zipf Mystery BigML Targeting Broad Queries in Search Project Mosaic Books How to eliminate bias in data driven marketing TensorFlow Dev Summit 2018 [videos] NLP Sentiment Analysis Talk 2 Books Image-Net The Shallowness of Google Translate TF-IDF LSI LDA Learn Python Massive Open Online Courses Coursera Machine Learning Advanced Resources
- 84. What did we learn?
- 85. What did we learn? ➢ Machine Learning combines statistics & programming ➢ A model is only as good as its training data ➢ The loss function helps us improve models, but overfitting should be avoided. ➢ YOU can create a ML model today!!! ➢ ML will help scale SEO tasks & allow us to evolve as SEOs
- 88. I lost it...
- 92. &
- 94. “One of the major promises of AI is freeing people from mindless tasks, so they can do more meaningful work.” -SAM DEBRULE
- 97. &
- 100. www.kdnuggets.com
- 102. Resource
- 103. Smart Compose:
- 104. &
- 105. Why SEOs Need To Know About ML: ML already effects the work that you’re doing. You should be able to speak intelligently about ML(especially with clients). Level up your skills by adding ML to your toolbox!
Editor's Notes
- #2: (35mins) We are drowning in data but starved for knowledge
- #6: Want to plant the seed of ML into all of your heads, dymistify it and show how accessible it is to eversingle person in this room
- #7: I wanna thank my parents for somehow raising me to have confidence that is disproportionate with my looks and abilities.
- #8: CodeLabs.
- #9: Inception model for flowers that I piggybacked off of. The best artists steal! <--That’s how I learn. Ok this works, how can I tweak it for what I need it to do?
- #11: Over 14 million images have been hand-annotated by ImageNet to indicate what objects are pictured
- #15: Bonus material I’ve removed from my deck
- #20: Who here enjoys adding alt text to images!? NO ONE!!! WE SHOULD BE SPENDING OUR TIME ON MORE IMPORTANT THINGS LIKE STRATEGY!
- #21: Who here enjoys adding alt text to images!? NO ONE!!! WE SHOULD BE SPENDING OUR TIME ON MORE IMPORTANT THINGS LIKE STRATEGY!
- #22: ML models just like that are already all around us. And they are saving lives!
- #24: Forget NLP, we are moving on to NLG
- #26: I aim to demystify Machine Learning and make it approachable for every single person in this room.
- #27: It’s funny we actually created a ML model at Moz to see if it could train itself to determine a better DA score than what we currently have. It discovered that
- #29: She criticised the Analytical Engine which was designed in the 1830s to use mechanical operations to automated certain computations from the field of mathematical analysis. She essentially noted how the AE assists in making things available that we’re already aware of. Turing Machine allowed us to decipher german encryptions to help us
- #30: We will be focusing on Supervised Learning
- #35: Clarify labels & importance of training data.
- #36: Just like we do, practice!
- #37: Mean squared error represents a common loss function that tells you how far off you are from perfect. While you likely won’t create a perfect model, you can continue making tweaks by continuing to run epochs & steps to get a closer fit. Vanishing gradient and overfitting are problems if you run the model too far.
- #38: The problem becomes overfitting, which does a poor job of accounting for new data. It’s essentially the same as:
- #47: First glance at keyword research FOR VOICE!!!!
- #50: Hey Google, talk to SEO quiz $200 adwords & a free tshirt!
- #51: So should we
- #52: Last year I talked about I helped improve Moz’s blog recommendations by changing the suggested articles to be pulled from just the first category.
- #57: Can be a bit depressing to show people the best stuff up front (unless you’re a search engine). Shops show delightful surprises throughout, websites can do the same thing and engagement can go up! But it’s a UX decision.
- #59: Large sites struggle to write meta descriptions at scale!
- #60: Large sites struggle to write meta descriptions at scale! ADD META RIGHT UNDERNETH
- #61: There are already models programmed to summarize long form content.
- #62: Took this person’s question and the top voted answer, plugged it into the model ...
- #63: NOT BAD!?
- #64: AUTOMATE THIS AT SCALE. FEATURED SNIPPET TERRITORY!
- #72: We are just scratching the surface of what’s capable
- #74: What will you come up with?
- #78: Don’t have to install tensorflow or other programs Free GPU Collaborative
- #79: Connect it back to SEO
- #80: Bias in data and the ethics of Machine Learning deserve an entirely other talk
- #82: Hardware is quickly improving following Moore’s Law
- #92: Netflix Kaggle story if I have time
- #93: Rand is my muse. Love this analysis on things he’s been negative vs positive about recently :)
- #97: Open source technology to break sentences down into components to better understand the meaning of words.
- #101: TensorFlow was originally developed by researchers and engineers working on the Google Brain Team within Google’s Machine Intelligence research organization.
- #106: By the end of this talk ((if I’ve done a good job) you will have these understandings covered.