| CPC G06F 16/24534 (2019.01) [G06F 16/2455 (2019.01)] | 17 Claims |
|
1. A computer-implemented method comprising:
for each item in a set of items in a data repository, deriving for the item an enhanced item embedding by i) determining property values of the item, ii) deriving a vector representation of the property values of the item, iii) deriving an item embedding for the item and iv) enhancing the item embedding by concatenating the item embedding with the vector representation of the property values of the item;
receiving a query and one or more filtering parameters; and
processing the query and the one or more filtering parameters by i) deriving a query embedding for the query, ii) deriving a vector representation of the one or more filtering parameters, iii) concatenating the query embedding with the vector representation of the one or more filtering parameters to derive an enhanced query embedding, and iv) performing a “k” approximate nearest neighbor search to identify “k” items in the data repository having property values satisfying the one or more filtering parameters;
wherein, in accordance with an encoding scheme for the set of items in the data repository, the value resulting from calculating the inner product of the enhanced query embedding and the enhanced item embedding of an item increases by a multiple of a predetermined fixed value “c” when the property values of an item satisfy the one or more filtering parameters received with the query, as compared to the value resulting from the calculation of the inner product of the query embedding and the item embedding.
|