公开(公告)号：US20200065669A1

公开(公告)日：2020-02-27

申请号：US16291115

申请日：2019-03-04

Applicant: Hitachi, Ltd.

Inventor： Yuya OKADOME ,  Wenpeng WEI ,  Toshiko AIZONO

IPC: G06N3/08 ,  G06N3/04 ,  G06Q30/06

Abstract: An error in a demand prediction of items is reduced. A data processing apparatus for processing data on a demand for multiple items includes a model learning unit that generates a prediction model for predicting the demand for the multiple items, and a demand prediction unit that predicts the demand for the multiple items by using the prediction model. The model learning unit inputs actual data of the demand for each item to a first neural network of each item, extracts a feature quantity of each item, and combines the feature quantity of each item to generate a second neural network that is a prediction model.

公开(公告)号：US20190066131A1

公开(公告)日：2019-02-28

申请号：US16081155

申请日：2017-03-15

Applicant: Hitachi, Ltd.

Inventor： Toshiko AIZONO ,  Wenpeng WEI ,  Takuya MOGAWA ,  Kazuaki TOKUNAGA ,  Yasuharu NAMBA

IPC: G06Q30/02 ,  G06N5/00

Abstract: When multiple explanatory variables are automatically created, a huge amount of output suggestions causes a heavy burden on selecting the suggestion. A marketing support system is configured to include a suggestion extraction unit that accepts purchase data and analyzes a correlation between the purchase data to output a composite variable, a restriction filtering unit that accepts the composite variable and a restriction table to exclude the composite variable based on a restriction condition defined in the restriction table, and a result filtering unit that uses a measure result defined in the past to estimate an anticipated effect when a measure based on the composite variable is performed, and selects a plurality of explanatory variables.

公开(公告)号：US20180247248A1

公开(公告)日：2018-08-30

申请号：US15758235

申请日：2015-10-28

Applicant: Hitachi, Ltd.

Inventor： Wenpeng WEI ,  Toshiko AIZONO

IPC: G06Q10/06

CPC classification number: G06Q10/06393

Abstract: For the purpose of reducing the processing amount in evaluating an effect of a measure with a KPI, a measure evaluation system is configured to have: a basic indicator calculation module grouping data into a subject and a non-subject and calculating basic indicators of the subject and the non-subject before measure execution; a basic indicator estimation module creating, for each basic indicator, an estimation model representing a relationship between the basic indicators of the subject and the non-subject and estimating an estimated basic indicator, which is exhibited when the measure is not executed on the subject, based on a basic indicator of the subject after measure execution and on the estimation model; and a KPI evaluation calculation module receiving a KPI definition formed of arithmetic calculation of basic indicators and calculating an estimated KPI value corresponding to the KPI definition with the KPI definition and the estimated basic indicator.