<p>Joint entity and relation extraction, as an important task in the field of natural language processing, has the core goal of synchronizing and accurately recognizing entity information in text, mining and determining the association relations between entities. However, existing joint extraction methods mainly face challenges such as semantic ambiguity caused by overlaps and redundancies of relation instances, and loss of semantic information due to insufficient modeling of local context. To address this issue, we propose BiRLA, a joint extraction model based on bidirectional semantic enhancement. We construct a contextual semantic understanding layer, build a hierarchical contextual representation via bidirectional long short-term memory network, and establish global semantic associations using a multi-head attention mechanism; A simple but effective decoding process is then ensured by a score-based classifier. The experiments show that compared with the existing mainstream models, the F1 values of the proposed model on the four public datasets NYT*, WebNLG*, NYT, and WebNLG are 93.3%, 94.5% , 93.2%, and 91.8%, respectively, which are all improved compared with the baseline model, verifying the effectiveness of the model.</p>