The joint extraction of entities and relations constitutes a critical component within the framework of information extraction. This process integrates named entity recognition (NER) along with relation extraction (RE). Although current joint extraction methods have made some progress, but they still face challenges like feature misclassification and limited interplay across the subtasks. To tackle these problems, we introduce a Convolutional Shared Feature Network for Joint Entity and Relation Extraction(CSFJRE). This method aims to establish a shared feature module to enable bidirectional information communication among entities and relations. The shared feature module can use entity information to assist in RE and leverage relationship information to aid in NER. Consequently, the tasks of NER and RE can mutually promote and enhance each other. Additionally, the CSFJRE model employs parallel encoding to obtain named entity features and relation features independently, effectively preventing feature overlap and misclassification. We evaluate our method in comparison to existing joint models using five benchmark datasets (NYT, WebNLG, ACE04, ACE05, and SciERC), demonstrating competitive performance.