Machine Learning (ML)

Machine learning (ML) is the branch of artificial intelligence in which systems learn patterns from data and use them to make predictions or decisions, instead of following rules a person wrote by hand. For SEOs and marketers it is the practical toolkit for classifying, clustering, extracting and predicting at a scale no spreadsheet of IF statements can manage.

How it works

A model is trained on examples, evaluated on data it has not seen, then used on new data. What kind of learning you need depends on your data:

• Supervised learning uses labelled examples to predict a category (classification) or a number (regression).
• Unsupervised learning finds structure in unlabelled data, such as clusters or topics.
• Reinforcement learning learns by trial and error towards a goal.
• Deep learning uses many-layered neural networks and can power both supervised and unsupervised tasks.

MLforSEO frames every project with three questions: what is the task, what data do you have, and what does the solution need to look like (for example, must it be explainable)? A useful first check, from Karen Hao’s flowchart: is the system finding patterns in large amounts of data? If not, it is probably a rule-based system, not ML.

How it’s used in SEO and marketing

• Search intent and text classification.
• Keyword and page clustering, and topic modelling.
• Entity extraction and sentiment analysis with pre-trained APIs.
• Fuzzy matching for redirect mapping and data clean-up.
• Forecasting and anomaly detection on traffic data.

Example

Matching the method to the task: mapping 404 URLs to live pages is a string-similarity problem, so fuzzy matching fits. Grouping customers by behaviour with no labels is clustering. Labelling queries as transactional when you have a labelled sample is supervised classification. Ranking a page by how much new information it adds draws on information gain, a concept that comes from decision trees.

Limitations and things to watch

• Output quality depends on input data quality.
• Many models give slightly different results each run, so build in checks for anything a client must be able to reproduce.
• Complex models are hard to explain to stakeholders.
• Not every problem needs ML: a well-designed rule or regex is often faster, cheaper and easier to maintain.
• Models drift as language, SERPs and products change, so re-test periodically.

Learn it on MLforSEO

• The Characteristics of a Machine Learning Problem: A Task, Data and Solution Framework for SEOs
• How to Find Machine-Learning-Enabled Automation in Your SEO Workflow
• What Machine Learning Classification Is, and Where It Fits in SEO
• When to use Machine Learning in SEO: Decision Checklist
• Introduction to Machine Learning for SEO (course)