Inferring　causal　effects　and　evaluating　decision　effects　has　attracted　widespread　attention　in　various　fields.　Under　the　framework　of　potential　outcomes,　unknown　counterfactual　outcomes　and　confounding　factors　are　obstacles　to　making　good　individual　treatment　effect　estimates.　And　most　work　assumes　that　the　obtained　covariates　are　all　confounding　factors,　while　it　is　difficult　to　guarantee　in　observational　data.　In　response　to　the　above　issues,　this　article　proposes　a　multitask　learning　framework　called　Decomposed-Representation　based　Causality　Estimating　Model　(DRCEM),　which　identifies　confounding　factors　by　learning　the　decomposition　representation　of　confounding　and　non-confounding　factors,　and　ultimately　estimates　the　causal　effect　in　the　observed　data　through　a　phased　multitask　training　process.　The　experimental　results　on　two　public　datasets　show　that　our　method　performs　well　and　outperforms　existing　causal　effect　estimation　models.　©　2024　ACM.