Read this paper for more details:
http://psiexp.ss.uci.edu/research/papers/SteyversShiffrinNelsonFormatted.pdf
WAS_BIGS2SOLUTION.mat
=====================
Contains SVD solution to the full 5018 x 5018 word association dataset. Take cosine similarity between vectors.
The SVD is based on S(2)_ij = S(1)_ij + sum_k[ S(1)_ik * S(1)_kj ] where S(1)_ij = A_ij + A_ji and A_ij is the
associative strength between item i and j.
N 1x1 8 double --> number of words
V 5018x400 16057600 double --> V( i , : ) contains the vector for word i (not normalized).
W 1x5018 623076 cell --> W{ i } contains word i
WAS_S2.mat
==========
Contains SVD solution to a subset of 2500 x 2500 word association dataset. Take cosine similarity between vectors.
The SVD is based on S(2)_ij = S(1)_ij + sum_k[ S(1)_ik * S(1)_kj ] where S(1)_ij = A_ij + A_ji and A_ij is the
associative strength between item i and j.
N 1x1 8 double --> number of words
V 2500x500 10000000 double --> V( i , : ) contains the vector for word i (not normalized).
W 1x2500 307642 cell --> W{ i } contains word i
WAS_MDS.mat
===========
Contains MDS solution to a subset of 2500 x 2500 word association dataset. Take Euclidean distance between vectors.
The MDS is based on distances T_ij = -log[ S(1)_ik * S(1)_kl * ... * S(1)_qj ] where S(1)_ij = A_ij + A_ji and A_ij is the
associative strength between item i and j.
N 1x1 8 double --> number of words
V 2500x500 10000000 double --> V( i , : ) contains the vector for word i (not normalized).
W 1x2500 307642 cell --> W{ i } contains word i