In order to characterize algorithmic trust in deep learning recommendation systems, this paper proposes a quantum decision trust modeling framework that integrates interaction representation, trust variable construction, quantum state evolution and recommendation response. A dataset consisting of 52418 users, 1.86 million clicks, ratings, dwell time and skip records, and 214730 trust feedback labels was constructed based on the online content platform. The sequence encoding structure is used to characterize the interaction sequence, and the quantum probability amplitude is used to model the trust superposition, interference and state transition in the recommendation exposure process. Then, a trust-aware ranking module was introduced to couple trust status and item relevance. Experiments on the MovieLens-1M, KuaiRec and self-built Trust-Rec datasets show that the proposed model achieves 94.8% Trust state recognition accuracy and 0.912 AUC on the Trust-Rec test set, and the average NDCG@10 reaches 0.242. The average recommendation acceptance rate reaches 73.7%, and stably converges after the 18th round. This framework provides an interpretable computational representation of cognitive fluctuations in recommendation scenarios.
