deprecatedabstract class %iKnow.Classification.Methods.VectorSpace
extends Base
This Builder Method implementation generates category term weights as a vector per category that
exists in the same vector space as the document term vector. The similarity between the
document vector and each of the category vectors can then be calculated using euclidean distance
or cosine similarity (angle). Alternatively, these weights can be used for a linear regression
formula, calculating a score rather than a similarity/distance.
A combination of global term weights (across the corpus), local term weights (within each
category) and normalization (per category) is used to calculate these category vectors.
• property CategoryGlobalTermWeights
as %String(VALUELIST=",none,IDF") [ InitialExpression = "IDF" ];
The corpus-wide relevancy factor to take into account when calculating term weights
• property CategoryLocalTermMetric
as %String(VALUELIST=",frequency,spread") [ InitialExpression = "spread" ];
The metric to use for calculating the local term weights
• property CategoryLocalTermWeights
as %String(VALUELIST=",binary,linear,logarithmic") [ InitialExpression = "linear" ];
The per-category term weight factor
• property CategoryNormalization
as %String(VALUELIST=",none,cosine") [ InitialExpression = "none" ];
Whether and how to normalize the category vectors
• property CustomTermWeights
[ MultiDimensional ];
