Latent Semantic Indexing

This lecture introduces Latent Semantic Indexing (LSI) as a method to improve information retrieval by mapping documents and queries into a lower-dimensional space of concepts, addressing issues like synonymy and homonymy. The instructor explains the key idea of LSI, the process of Singular Value Decomposition (SVD), and how to identify top concepts. Through examples and illustrations, the lecture covers the construction and interpretation of SVD, similarity computation in the concept space, and answering queries using cosine similarity. The lecture concludes with practical examples of applying LSI to term-document matrices and mapping queries into the concept space.