Topic Modelling Service: Discover What Your Documents Are About
When you have thousands of texts and no time to read them all, topic modelling shows what they are about. We discover topics, label them sensibly and show how they change across time and groups.
What this service covers
- Text preparation: cleaning, tokenisation, stop words and phrases
- Topic models: LDA, NMF and embedding-based approaches such as BERTopic
- Choosing the number of topics with coherence scores and human review
- Topic labels, top words and representative documents
- Topic prevalence over time and across groups
- Interactive visualisations of topics
- A written interpretation tied to your research question
What topic modelling is and is not
Topic modelling groups texts that use words in similar ways. A topic is a pattern of co-occurring words, not a guaranteed human concept, so labelling and checking is essential. We read representative documents for each topic, adjust the model when topics are incoherent and report how stable the results are.
Choosing the method
LDA works well with longer documents and is easy to explain. For short texts such as tweets or survey comments, embedding-based methods like BERTopic often give clearer topics. We try suitable methods, compare them on coherence and usefulness, and recommend one. When you need rigorous interpretation of a smaller set, thematic analysis may be preferable, and we can combine the two.
Showing change
Often the interesting question is how topics rise and fall, or which groups talk about what. We chart topic prevalence across time, sources or segments, and pair it with sentiment analysis to show how people feel about each topic. Try the idea on a small sample with our free theme finder.
Related services
See text classification for assigning texts to known categories and entity extraction and document analysis for large collections.
How the work runs
- You share the corpus and your questions.
- We prepare the text and test models.
- We label, check and visualise topics.
- You receive the results, charts and interpretation.
What you receive
- A labelled topic list with top words
- Representative documents for each topic
- Topic trends over time or by group
- Interactive or static visualisations
- A written interpretation
Frequently asked questions
How many topics should there be?
It depends on the data and purpose. We compare options using coherence and manual review rather than guessing.
How much text do I need?
Hundreds of documents may work for broad topics; thousands are better for fine-grained ones.
Can topic modelling work on short texts?
Yes, with methods designed for short text, though results need careful checking.
Is LDA outdated?
No, it is still useful and interpretable. Newer embedding-based methods can be better in some cases.
Can you model texts in other languages?
For many languages, yes. Tell us the language.
Can you combine this with qualitative coding?
Yes. Topic modelling can guide a human coding frame, and coding can validate topics.
Related nlp & text analytics services
Sentiment Analysis
Positive, negative and neutral sentiment for reviews, comments and posts.
Text Classification
Categorise comments, tickets and documents into your own labels.
Keyword & Word Frequency Analysis
Text cleaning, n-grams, keywords in context and word clouds.
NLP & Text Analytics Services
See every text analytics service we offer.
Also related: R, Python and MATLAB analysis