| CPC G06F 16/24578 (2019.01) [G06F 16/24542 (2019.01); G06F 16/2455 (2019.01)] | 20 Claims |
|
1. A method, comprising:
at a computer system comprising at least one processor and memory:
receiving a plurality of search queries from a plurality of users directed at an item query engine of an online system, each search query including a search phrase used by a user to conduct the search query;
generating a training set comprising a plurality of training samples, the plurality of training samples comprising a plurality of search phrases, each training sample corresponding to a search query of the item query engine and comprising a search phrase and a list of items returned by the item query engine;
generating a representative set of training samples, wherein generating the representative set of training samples comprises:
determining search frequencies of the search phrases used in the training samples,
stratifying the set of training samples into a plurality of bins according to the search frequencies of the search phrases, wherein each bin of the plurality of bins includes a subset of training samples, wherein each bin defines a range of numbers of times a search phrase is used by the plurality of users, and
sampling the training samples from the plurality of bins to collect the representative set of training samples; and
training the item query engine using the representative set of training samples.
|