Figure 3

Varying the dimensionality of the index vectors shows the effect of varying the dimensionality (k) of the random index vectors. The ratio of non-zero elements is kept fixed, varying from 2 (for k=1,250) to 32 (for k=20,000). The plot shows averaged and maximum sentence-level F₁ across 5 different runs for each setting (using different random initializations of the index vectors), testing on 1/10th of the training data. For reference, the last column shows the result for using the original non-projected feature space of 670,000 dimensions.