Named Entity Recognition of E-commerce Texts Fusing Large Language Models and Lexical Knowledge

Aiming at the problem that Chinese e-commerce text data set covers many fields and various kinds, this paper constructs a synonym table combined with the powerful information search and learning ability of large language model, and carries out data enhancement by randomly replacing the synonym text in the text of the data set, which makes the data set more rich. In addition, a LE-Roberta-wwm-BiLSTM-CRF model integrating vocabulary information is proposed. First, word information related to each word vector is fused into the pretraining model Roberta-wwm-ext-large, so that full and effective learning and training can be carried out in the pre-training stage. The feature is extracted by BiLSTM, and the conditional random field model (CRF) is usedto calculate the conditional probability and optimize the output, so as to improve the accuracy of e- commerce text named entity recognition. The experimental results show that the comprehensive accuracy, recall rate and F1 of the proposed method are 90.64%, 92.11% and 91.36%, respectively, which proves that the proposed method can better improve the named entity recognition accuracy of Chinese e-commerce texts.