IBM Watson NLU (Natural Language Understanding)

IBM Watson Natural Language Understanding (NLU) is a cloud API that analyses text and returns structured data about it: entities, keywords, sentiment, emotions, relationships between entities, categories and more. It is one of the three purpose-built NLP APIs MLforSEO compares, alongside the Google Cloud Natural Language API and Amazon Comprehend, and it stands out for the range of features it returns from a single call.

How it works

You send text, HTML or a URL to the analyze endpoint and choose which features you want:

• Entities with type, relevance, mention count, confidence and, for some, a DBpedia link; optional per-entity sentiment and emotion.
• Keywords with relevance, plus optional sentiment and emotion.
• Sentiment for the document or for target phrases, labelled positive, negative or neutral with a score.
• Emotion: joy, sadness, fear, disgust and anger scores.
• Relations between entities (for example, “employedBy” or “locatedAt”) and semantic roles (subject, action, object).
• Categories in a hierarchical taxonomy, concepts, syntax and page metadata.

Status. IBM deprecated Watson NLU on 31 July 2026 and plans to remove it from IBM Cloud on 31 January 2028, recommending migration to Watson Natural Language Processing in watsonx.ai. The summarisation feature was already retired in October 2024.

How it’s used in SEO and marketing

• Review and feedback analysis that goes beyond positive or negative into specific emotions.
• Mapping relationships between entities in content, useful for knowledge graph and entity SEO work.
• Keyword and concept extraction from competitor pages.
• Categorising content with a ready-made taxonomy.

Example

In MLforSEO’s benchmark, the same 221-word bio was sent to each API. IBM returned 16 entity rows (15 unique), found 7 of the 13 core named entities in full, and made the most visible errors, such as splitting “Women in Marketing – Bulgaria” into a job title and a location. Where IBM shone was the extra data per entity (sentiment, relevance, count, confidence and DBpedia disambiguation) and its relations and emotion modules, which the other APIs do not offer in the same way.

Limitations and things to watch

• With removal scheduled, do not build new long-term pipelines on it; plan a migration for existing ones.
• Entity recall was weaker than Google’s in MLforSEO’s test, so check outputs on your own text.
• Emotion and sentiment scores are model estimates and APIs often disagree; validate on a labelled sample.
• Feature availability varies by language.

Learn it on MLforSEO

• Comparing NLP APIs for Entity and Sentiment Analysis (vs Generative AI)
• IBM Watson NLU Complete API Implementation Template (Notebook)
• Entity Analysis & Relationship Mapping with IBM Watson NLU in Google Colab
• IBM Watson NLU Sentiment and Emotion Analysis Template for Google Sheets
• Introduction to Machine Learning for SEO course