Glossary

Topic modeling

An unsupervised machine learning technique that discovers the latent themes running through a collection of documents.

Topic modeling is an unsupervised machine learning technique that scans a large collection of documents and identifies the latent themes, or topics, running through them, without being told in advance what those topics are.

Classic approaches such as Latent Dirichlet Allocation represent each document as a mixture of topics and each topic as a distribution over words, inferring both simultaneously from patterns of word co-occurrence; newer approaches instead cluster document embeddings produced by language models. This differs from Named Entity Recognition, which extracts specific known entities: topic modeling surfaces unnamed, data-driven themes that a human then has to label.

Analysts use topic modeling to summarize large volumes of open-ended survey responses, support tickets, or social posts collected through social listening, turning unstructured text into a manageable set of themes for reporting. It is closely related to clustering and unsupervised learning more broadly. The main pitfall is that the number of topics is usually chosen by the analyst rather than discovered automatically, and topics can be unstable or hard to interpret, requiring manual labeling and validation before results are trusted.

Last reviewed September 22, 2026

In the index now

Related terms

Text analytics The techniques used to extract structured information and insight from unstructured text, from keyword counts to deep NLP. Named Entity Recognition (NER) An NLP technique that locates and classifies names of people, organizations, places, and other entities within text. Sentiment analysis Using NLP to classify text as positive, negative, or neutral, so large volumes of comments or reviews can be scored automatically. Clustering An unsupervised technique that groups data points so items within a group are more similar to each other than to items outside it. Unsupervised learning A machine learning approach that finds structure or patterns in data that has no labeled correct answer to learn from.

Related tools

Gensim Open-source Python library for unsupervised topic modeling and document-similarity analysis using vector-space methods.

Related guides

How to analyze text data From word counts and topic models to LLM classification — matching the method to the question instead of reaching for the newest tool by default. How to choose a digital humanities tool Digital humanities tools split by what they analyze — text, networks, images or structured historical data — pick by your source material, not the category. How to choose a text analytics or NLP tool Pick an NLP library or platform by whether your team writes code, what task you're actually doing, and whether you need a pretrained model or a custom one.