The goal of Dialogue State Tracking (DST) is to track the user’s intents and goals in each conversation round based on a given dialogue history. Existing DST approaches generally predict dialogue states at each turn with semantic information from dialogue history and query slots. However, most of them neglect the dialogue acts which reflect speakers’ intents and do not consider the different effects of dialogue acts on categorical slots and non-categorical slots. To leverage the semantic information from dialogue acts, we propose a novel and effective model called IAAD-DST which incorporates both user acts and system acts for state prediction. In particular, to model the interaction between user and system, an interactive acts attention method is developed to extract the interactive information between user acts and system acts within a turn. Besides, an attention divergence loss function is utilized to train model to extract acts feature with high matching degree to provide more precise acts feature for span prediction. Experimental results on the MultiWOZ 2.1 and MultiWOZ 2.2 datasets demonstrate that our model performers well and outperforms baseline models, demonstrating the effectiveness and necessity of incorporating dialogue acts.