To help candidates understand the specific needs of the market for talents, and provide references for education institutions to develop training programs and curricula. Scraping the recruitment data on the 51Job website with statistics as the keyword. After cleaning and labeling them, comparing the entity recognition effects of BiLSTM-CRF, BiGRU-CRF, BERT-BiLSTM-CRF, and BERT-BiGRU-CRF, and then selecting the best model to extract recruitment entities for mining analysis. The experimental results show that the BERT-BiGRU-CRF model has the best recognition effect, and the accuracy value and F1 value are increased by 7.05% and 4% respectively, which has certain advantages over the other three models. According to the selected recruitment entities, the skills and qualities that the statistics profession should possess are summarized from the three aspects of educational background, professional requirements, and personal qualities, and suggestions on training for statistics education are put forward.