Multi-scenario recommendation algorithms are widely adopted in online applications. While current gated expert network approaches improve scene extraction, they suffer from over-reliance on supervised signals and poor sparse-scenario performance. Existing contrastive learning methods also face challenges in sample construction due to dependency on predefined feature labels. We propose DCPRec, a Dual-Parameter Contrastive Recommendation model that integrates enhanced Transformer encoding with bidirectional parametric networks for cross-scenario modeling. Through novel dual-view contrastive learning, DCPRec achieves robust feature enhancement. Evaluations on Nowcoder’s clickstream data demonstrate superior performance in interpretability and transferability, with successful deployment in Nowcoder’s article recommendation system showing significant improvements across real-time recommendation scenarios.