Hu ; X.1 ; Cai ; Z.1 ; Franceschetti ; D.2 ; Penumatsa ; P.1 ; Graesser ; A. C.1 ; Louwerse ; M. M.1 ; McNamara ; D. S.1 ; the Tutoring Research Group1
We report two discoveries concerning Latent Se mantic Analysis (LSA). First, we observed the special properties of the …rst dimension of the LSA space. Second, we observed that dimensional weighting plays an important role in LSA analysis. Based on the …rst discovery, we examined the cosine matches without the …rst dimension. Based on the second discovery, we explored dierent dimensional weighting schemes. Based on these observations, we recommend a new algorithm for LSA cosine com putation such that LSA becomes more sensitive to