Question

For any two documents x and z, define k(x, z) to equal the number of unique words that occur in both x and z (i.e., the size of the intersection of the sets of words in the two documents). Is this function a kernel?

Answer 1

Answer:

Yes, the function k(x,z) is a kernel

Explanation:

To show that k(x,z) is a kernel.

First, we'll need to explicitly construct two features of k(x,z).

The features are:

vector φ(x) and vector φ(z) in such a way that K(x,z) =φ(x)·φ(z).

Then, we proceed to the next step...

For any given documents, a vocabulary V can be constructed.

Vocabulary V is finite size for the words in the document set.

Given V, a feature mapping φ(x) can then be constructed for x by the following:

For the kth word wk in V, if wk appears in document x, assign φ(x) k(the kth element of φ(x)) to be 1; else assign 0 to it.

Then the number of unique words common in x and z is φ(x)·φ(z).

This gives us the kernel.