Mastering Topic Modeling, Anomaly Detection, and PageRank
Topic Modeling: LSI vs LDA
Latent Semantic Indexing (LSI)
Use LSI for semantic similarity, retrieval, short/sparse documents, and synonymy problems. It uses Singular Value Decomposition (SVD) to find latent concept axes from term-document co-occurrence. Concepts are mathematical directions, not clean probability-based topics. It is better when the goal is to “find similar documents.”
Latent Dirichlet Allocation (LDA)
Use LDA for discovering hidden themes and topic percentages. In this model, each document is a mixture of topics, and each topic is a distribution over words. It is better when the goal is to determine “what themes exist in this corpus?”
Key Distinction
- LSI finds latent concept dimensions.
- LDA explicitly models probabilistic topics.
English with a size of 12.48 KB
