Natural Language Understanding (NLU)

Natural language understanding (NLU) is the branch of natural language processing concerned with comprehension: working out the meaning, intent and context of text so a machine can act on it. Where NLP covers every step of handling language, from tokenisation to generation, NLU focuses on interpretation, such as what a customer wants, which entities a review mentions, and how the writer feels about them.

How it works

NLU combines several analyses, which commercial APIs usually expose as separate features:

• Syntactic analysis: parts of speech and grammatical structure.
• Entity recognition and linking: finding named things and matching them to a knowledge base.
• Sentiment, emotion and entity sentiment: how the writer feels overall and about each entity.
• Intent detection and classification: what the user wants, or which category the text belongs to.
• Relation extraction: how entities in the text relate to each other.

IBM sells an API under the name Watson Natural Language Understanding, with modules for entities, keywords, sentiment, emotion, categories and relations. Google Cloud Natural Language and Amazon Comprehend cover similar ground. Large language models now perform many NLU tasks from a prompt, with different trade-offs in consistency and cost.

How it’s used in SEO and marketing

• Query understanding: classifying search intent and extracting entities from keyword lists.
• Customer reviews: combining sentiment, emotion and entity analysis to find what customers praise or complain about.
• Content audits: checking which entities a page is understood to be about.
• Chatbots and support: routing messages by intent.

Example

In MLforSEO’s semantic analysis of customer reviews, reviews are sent to both Google Cloud Natural Language and IBM Watson NLU. Google returns entity sentiment; Watson adds emotion scores such as joy, sadness and anger. Together they show, for instance, that “delivery” is mentioned often with negative sentiment while “fit” is associated with joy, which gives product and content teams specific issues to address.

Limitations and things to watch

• Sarcasm, domain jargon and mixed sentiment still trip up NLU systems.
• Different APIs return different entities and scores for the same text, as MLforSEO’s API comparison shows, so choose one per project and stay consistent.
• Language coverage varies by feature and provider.
• API outputs are probabilistic estimates; validate a sample by hand.

Learn it on MLforSEO

• Semantic Analysis of Customer Reviews with Machine Learning
• Comparing NLP APIs for Entity and Sentiment Analysis (vs Generative AI)
• IBM Watson NLU Complete API Implementation Template for Multi-Feature Text Analysis (Notebook)
• What Entity Extraction Is, and Where It Fits in SEO
• Introduction to Machine Learning for SEO (course)