Transfer learning model for estimating happiness from historical purchasing data and questionnaire data

Abstract

Although there exist some big-data that are collected at reasonable cost easily, many other data take much cost to collect them. One of the latter cases includes historical purchasing and questionnaire data with IDs, simultaneously. For such case, we can often use only historical purchasing data for other IDs, additionally, however such data do not include questionnaire data.
As for applying questionnaire data for other customers, transfer learning model can be a promising approach, and we propose a transfer learning model to estimate happiness for a practical fashion site data. From some computational experiments, we show practical results and discuss about them.