Topic Modeling is an unsupervised method for document classification (classify into topics the model invents by itself), aiming to discover and extract hidden thematic structures within large collections of text data.
We can achieve this by analysing word co-occurance patterns within unstructured text data. The hidden thematic structures can then be used to classify and summarise the available data.
Here are the common Topic Modeling techniques used:
- Latent Dirichlet Allocation (LDA)
- Latent Sementic Analysis (LSA)
- Probabilistic Latent Semantic Analysis (pLSA)
- Non-negative Matrix Factorisation (NMF)
All 4 methods squeeze a wide term matrix into a few topics, and the general form of that idea is Dimensionality Reduction.