Similarity Between Semantic Spaces
Zhiqiang Cai, Arthur C. Graesser, Xiangen Hu, Matthew Ventura · eScholarship (California Digital Library) · 2005
One of the challenges in Latent Semantic Analysis (LSA) is deciding which corpus is best for a speci¯c application.Imp ortant factors of LSA in°uence the generation of high quality LSA space including the size of the corpus, the weight (local or global) functions, number of dimensions to keep, etc.These factors are often di±cult to determine and as a result hard to control for.In this paper, we provide a general method to measure similarity between semantic spaces.Using this method, one can evaluate semantic spaces (such as LSA spaces) that are generated from di®erent sets of parameters or di®erent corpora.The method we have develop ed is generic enough to evaluate di®ering types of semantic spaces.