Question

for a task like recommending a item for a user, I can split dataset in two: traning and test datasets, and I can evaluate the result by using mean absolute error. However, when I need to find similar items of a item, I don’t know how to evaluate the results.for example,

input:

    userid,itemid,score
    1,      123,    5
    2,      222,    1
    ....

svd.similar(123)

the output maybe: (222,0.98),(121,0.78)…
I don’t know the similar items found from svd are good or not, how to evaluate the result?

Answer 1

You might want to define a similarity function (distance function) first, e.g., cosine distance/euclidean distance. Then similarity can be determined and compared.