<p>Multi-View Clustering (MVC) is crucial for multi-source data integration in intelligent information systems, where robust representations are required for recommendation, user profiling, and multimedia understanding. In real-world scenarios, however, views are often missing because of sensor failures, transmission interruptions, privacy constraints, or asynchronous collection. Such missingness degrades both feature completeness and cross-view structural consistency, causing completion noise to propagate to alignment and clustering. To address this challenge, we propose SDSWC-IMVC. Specifically, we construct k-nearest-neighbor (KNN) neighborhoods in each view, estimate cross-view structural similarity via Jaccard overlap on co-observed samples, and select the most structurally consistent source view for missing-view completion. To preserve geometry during optimization, we combine neighborhood locking for observed samples with dynamic neighborhood updating for missing samples under implicit graph Laplacian regularization. We further introduce prototype-based weak alignment to complement instance-level strong alignment, improving semantic stability while preserving discrimination. By jointly optimizing reconstruction, structural regularization, and strong-weak contrastive objectives, our model forms a structure-semantics closed loop. Experiments on four benchmark datasets demonstrate superior robustness across missing rates, indicating strong practical value for incomplete heterogeneous data in modern intelligent information and recommendation systems.</p>