Latent semantic mapping [information retrieval
J.R. Bellegarda · IEEE Signal Processing Magazine · 2005
This article has described LSM, a data-driven framework for modeling globally meaningful relationships implicit in large volumes of data. LSM generalizes a paradigm originally developed to capture hidden word patterns in a text document corpus. Over the past decade, this paradigm has proven effective in an increasing variety of fields, gradually spreading from query-based information retrieval to word clustering, document/topic clustering, large-vocabulary speech recognition language modeling, automated call routing, semantic inference for spoken interface control, and several other speech processing applications.