Pull request (PR) review plays a pivotal role in collaborative software development, yet the growing volume and complexity of contributions in large-scale open-source projects have made it increasingly difficult to assign suitable reviewers in a timely manner. Existing recommendation approaches primarily focus on either interaction histories or PR content, often neglecting structured semantic relationships between users, code changes, and their contextual attributes. Moreover, most methods fail to incorporate rich user-side information such as expertise, preferences, and behavioral activity into the recommendation process. In this paper, we propose DCKG, a Dual-View Collaborative Knowledge Graph framework for personalized PR recommendation. DCKG constructs a heterogeneous graph where both users and PRs are modeled as central nodes, enriched with semantic attributes including user skills, topic preferences, activeness levels, PR labels, file paths, and contributor identities. To bridge unstructured content and structured reasoning, we employ a large language model (LLM) to extract symbolic features from review comments, PR titles, and code diffs. Extensive experiments on a real-world PR dataset demonstrate that DCKG significantly outperforms strong baselines. Additional analyses show that incorporating user-side semantics improves interpretability and robustness in sparse interaction settings.