To address the following two major drawbacks of existing Contrastive Learning （CL） methods for the nested Named Entity Recognition （NER） tasks： 1） candidate entities by greedily enumerating in contrastive learning lack contextual semantics and boundary information， 2） unnecessary noise and invalid information increases computational burden and weakens contrastive learning performance， a two-stage NER framework was proposed. In the first stage， candidate entity boundaries were generated by the boundary recognition model， and candidate entities were integrated by the boundary integration module to minimize unnecessary negative candidates. Attention cues were inserted on both sides of the candidate entities to generate corresponding candidate entity texts， allowing the model to perceive contextual semantics and boundary information. In the second stage， a bi-encoder framework mapped candidate entity texts and entity label annotations into the same vector representation space through contrastive learning， with the comparison objects being sentences with attention cues rather than candidate entities. In addition， a classification parameter matrix with label semantics was designed to enrich the model’s understanding of candidate entities.Experimental results show that compared with Binder method， the proposed method improves the F1 values of 1.22， 3.42 and 2.31 percentage points， respectively， on three nested datasets： GENIA， ACE2005 and ACE2004， which verifies the effectiveness of the proposed method for tasks of nested NER.