machine learning - How to evaluate a suggestion system with relevant order? -
i'm working on suggestion system. given input, system outputs n suggestions.
we have collected data suggestions users like. example:
input1 - output11 output12 output13 input2 - output21 input3 - output31 output32 ...
we want evaluate our system based on data. first metric if these outputs present in suggestions of our system, that's easy.
but now, test how positioned these outputs in suggestions. have given outputs close first suggestions.
we single score system or each input.
based on previous data, here score of 100% be:
input1 - output11 output12 output13 other other other ... input2 - output21 other other other other other ... input3 - output31 output32 other other other other ... ...
(the order of output11 output12 output13 not relevant. important ideally 3 of them should in first 3 suggestions).
we give score each position hold suggestion or count displacement ideal position, don't see way this.
is there existing measure used ?
you want called mean average precision (it's metric information retrieval).
essentially, each of 'real' data points in output list, can compute precision (#of correct entries above point / #entries above point). if average number across positions of each of real data points in output list, metric want.
Comments
Post a Comment