Research on News Text Entity Recognition Technology Based on Deep Learning

With the rapid development of Internet technology, the volume of news text data has grown explosively, making it a key research focus to efficiently and accurately extract valuable information from this vast amount of news text. To address the challenges in entity recognition within news texts, this paper proposes a deep learning-based Named Entity Recognition (NER) method — the BERT-BiLSTM-MSConvAttn-CRF model. This model first uses the BERT pre-trained language model to obtain high-quality word embedding vectors, which can capture rich semantic and syntactic information of words in specific contexts. Then, these word embedding vectors are processed through the BiLSTM layer, effectively capturing long-distance dependencies in the text, thereby enhancing the understanding of sentence structure. To further improve the model's ability to identify key entity features, we introduce the MSConvAttn layer, which combines multi-scale spatial and channel information through a multi-scale convolutional attention mechanism, allowing for more precise capture of important details in the text. Finally, the CRF layer is used to globally optimize the entire sequence, ensuring that the output label sequence is both coherent and accurate. The proposed model was validated using a named entity recognition dataset for news texts, and experimental results show that the model outperforms existing models in NER tasks, providing a new solution for information extraction from news texts.