The BERT-BiLSTM-CRF Model Applied to Chinese Entity Recognition for the Science and Technology Service Field

Nowadays, the application of named entity recognition task just focuses on certain entities. When it comes to a specific vertical field, the scarcity and source of corpus become a problem. Science and technology resource service is such a vertical field without ready-made corpus. This problem has become a breakthrough point on its domestic development stage. The paper proposes a named entity recognition task which can be applied in the field of science and technology resource service, our data are based on the seeking for human resources which is a common business scenario in science and technology resource service. We use BERT model to generate word vector and combine BiLSTM's memory ability of context relationship with CRF's learning ability of annotation rules to build model. Moreover, the paper grabs the text information of recruitment websites with similar language organization structure, the same entity category and similar semantic expression, and carries out entity recognition for the “profession” entity and the “technology” entity. Finally, the model was trained in the produced data set to obtain a reasonable recognition effect, which proves the feasibility of named entity recognition in talent seeking, and can be further applied to the dialogue system of science and technology resources service to contribute to the development of its platform.