What does Multidimensional Scaling after LSA depict



In the below image a 3d semantic space has been created:

#The code for this:

fit <- cmdscale(dist.mat.lsa, eig = TRUE, k = 3)
colors <- rep(c("blue", "green", "red"), each = 3,length.out = length(fit$points[,1]))
scatterplot3d(fit$points[, 1], fit$points[, 2], fit$points[, 3], color = colors,
              pch = 16, main = "Semantic Space Scaled to 3D", xlab = "x", ylab = "y",
              zlab = "z", type = "h")

However I am not being able to interpret this graph.
Can someone please help me in understanding how to interpret it??


Hi @pagal_guy

I could be wrong but you display the three dimensions of the U matrix of the SVD here. As PCA you keep the dispersion of the the points in the cloud, it means that you have observations clubed together at x,y 0 and z about 5, the cosine should close to one for all those and therefore the document similar (represent by the two in you term to document matrix). You have few documents that are difference very obviously, in case of the point at x= -15, y=-30, z = -10 and the extreme green point at x=-30. In few words if you do text search pass x=-5 the documents should be different than the rest below this value, check the spear on y as well.
Hope this help.