Leiden University · Faculty of Humanities

Thesis & Research Supervision

Topic Analysis

Finding the themes that run through a corpus without reading every text — useful for exploratory work on large collections and for measuring how themes vary across time, source, or speaker.

What it is

Topic analysis is a family of techniques that discover clusters of co-occurring words (“topics”) across a corpus, along with the proportion of each topic in each document. The best-known variants are Latent Dirichlet Allocation (LDA) and its metadata-aware cousin the Structural Topic Model (STM); newer embedding-based methods like BERTopic occupy the same analytical slot with a different machinery. Outputs are usually (a) lists of high-probability words per topic, and (b) document-level topic proportions you can use for downstream comparison.

Topic analysis does not “find the truth” of a corpus — it finds statistical regularities you then have to interpret. Naming, validating, and defending the topics is the researcher’s job, and is where methodological rigour lives.

What you learn in the DH course

This page draws from the course’s topic modeling material. Students who take it come away with:

  • How LDA works conceptually (mixed-membership over topics; topics as word distributions)
  • Reading STM as LDA plus covariates that shift topic prevalence and content
  • Embedding-based topic methods (BERTopic, Top2Vec) and how they differ from LDA
  • Choosing K (the number of topics): diagnostic plots, coherence scores, interpretability checks
  • Validating topics: semantic coherence, intruder tests, human coding on a sample
  • Measuring topic prevalence across metadata covariates (year, source, speaker)
  • Reporting topic models in a methodology chapter — what has to be disclosed

What you need to learn first

  • Preprocessing: topic models are notoriously sensitive to preprocessing. See Preprocessing.
  • Basic statistics and probability: enough to understand “mixture over distributions” without being intimidated by it.
  • R or Python: STM is an R package; LDA and BERTopic have strong Python tooling (gensim, scikit-learn, bertopic).

What you can do with it

  • Track how themes in a news corpus shift across a political crisis (prevalence by month)
  • Compare how two political parties frame the same issue (prevalence by source)
  • Identify latent genres in a literary corpus that weren’t defined a priori
  • Surface candidate themes for subsequent close reading — topic modeling as a “where to look” tool rather than a final answer
  • Produce descriptive scaffolding for a larger corpus that would otherwise be impossible to read end-to-end
  • Preprocessing — the input to a topic model; shapes every topic you see.
  • Word Embeddings — embedding-based topic models (BERTopic) and contextual-embedding methods live here.
  • Framing Analysis — topics-as-frames is a common analytical move; qualitative framing analysis on a topic-model-selected subset is a powerful pairing.
  • Discourse Analysis — topic-modelled themes can drive sampling for close reading.