The objective of dialogue state tracking (DST) is to update the current dialogue state on the basis of the utterance information from the current turn and the dialogue state from the previous turn. Traditional methods treat the previous turn’s dialogue state as a holistic representation, frequently overlooking critical details from earlier turns and resulting in ’forgetting’ issues. To address this limitation, a discounted return algorithm is employed to aggregate dialogue history information more effectively. Additionally, to capture slot-specific information within utterances, we introduce an utterance-slot attention mechanism that can model the semantics of the dialogue context while incorporating slot name features to guide the generation or extraction of the dialogue state. The experimental results not only validate the superiority of our proposed method but also significantly increase the accuracy and robustness of multidomain DST.