Semantic Analysis

Semantic analysis is the part of natural language processing concerned with meaning. It works out what words, phrases and sentences mean, which things are being talked about, and how they relate to each other. Entity extraction, sentiment analysis, text classification and topic modelling all sit under this heading.

How it works

MLforSEO explains semantic analysis with an analogy from compiler design, where human language passes through a series of phases on its way to machine understanding:

1. Lexical analysis breaks text into tokens and word forms.
2. Syntax analysis works out grammatical structure.
3. Semantic analysis checks that the relationships make sense and uncovers what the words actually mean.

You rarely build this yourself. Pre-trained APIs such as the Google Cloud Natural Language API, IBM Watson NLU and Amazon Comprehend bundle the semantic tasks together: entities with types and salience, document and entity sentiment, classification and moderation. Large language models can perform the same tasks from a prompt, with less consistent output.

How it’s used in SEO and marketing

• Customer reviews. Extract entities, entity sentiment, emotion and moderation flags from reviews to see what customers love and what frustrates them.
• Query analysis. Extract entities from queries to classify intent and spot topics.
• Content audits. Find each page’s main entities and use them for internal linking and gap analysis.
• Topic discovery. Cluster content into themes with topic modelling.
• Keyword research. Build a semantic keyword universe organised by entities and meaning.

Example

Take the review “The staff in the Leeds store were lovely, but the prices went up again.” Semantic analysis finds the entities staff, Leeds store and prices. Entity sentiment is positive for staff and negative for prices, and the overall document sentiment is mixed. A plain positive-or-negative label would hide the fact that the store’s service is a strength and pricing is the problem.

Limitations and things to watch

• Different APIs return different entities and scores for the same text, so compare outputs before committing to a provider.
• Google’s Natural Language API v2 (2023) dropped salience scores and entity sentiment, which v1 still offers.
• Sarcasm, idioms and domain jargon reduce accuracy.
• Meaning is not the same as intent: knowing what a query is about still leaves the question of what the searcher wants to do.

Learn it on MLforSEO

• What Entity Extraction Is, and Where It Fits in SEO
• Semantic Analysis of Customer Reviews with Machine Learning
• Comparing NLP APIs for Entity and Sentiment Analysis
• Complete Guide to Entity Extraction and Sentiment Analysis
• Introduction to Machine Learning for SEO course