US 12,229,106 B2
Table integration system, method, and program
Yuyang Dong, Tokyo (JP); and Masafumi Oyamada, Tokyo (JP)
Assigned to NEC CORPORATION, Tokyo (JP)
Appl. No. 18/023,830
Filed by NEC Corporation, Tokyo (JP)
PCT Filed Sep. 2, 2020, PCT No. PCT/JP2020/033310
§ 371(c)(1), (2) Date Feb. 28, 2023,
PCT Pub. No. WO2022/049682, PCT Pub. Date Mar. 10, 2022.
Prior Publication US 2023/0237039 A1, Jul. 27, 2023
Int. Cl. G06F 16/22 (2019.01)
CPC G06F 16/2282 (2019.01) [G06F 16/221 (2019.01)] 9 Claims
OG exemplary drawing
 
1. A table integration system comprising:
a coupling table storage that stores a coupling table couplable to a base table including a target column including target data in a trained model that predicts sales of a product by learning;
a candidate column storage device that stores a candidate column including data correlated with the target data of the target column; and
an integrated table generation device that generates an integrated table obtained by coupling the base table and the candidate column,
wherein the integrated table generation device includes:
a memory storing instructions; and
one or more processors configured to execute the instructions to:
receive input of the coupling table, the candidate column, and the base table;
couple the candidate column in the coupling table to the base table;
train the model with data of a designated target column as an objective variable and with the data of the candidate column as one or features;
evaluate the trained model based on a predetermined evaluation method;
select, from among the one or more features, the feature that improves performance of the trained model based on the data included in the candidate column that has been coupled to the base table, by selecting the feature that improves an evaluation index calculated by the evaluation method; and
output the integrated table obtained by coupling a column including the selected feature and the base table.