Research on Data Mining Methods in the Field of Quality Problem Analysis Based on BERT Model

In the current quality problem analysis work, there are problems such as large amount of data with scattered distribution, isolated data and difficult machine understanding. Most of the current data mining work in the field is based on deep learning models, which is difficult to be integrated into the characteristics of the data in the field. Also, there are still some deficiencies in its accuracy rate and training speed. Therefore, this paper carries out the research of data mining methods in the field of quality problem analysis and incorporates the characteristics of data in the field on the basis of the existing model construction research. Focusing on the named entity recognition task and the relationship extraction task, the data are preprocessed by sequence annotation method to form the data sets of the two tasks. For the first task, a BERT-based recognition method is adopted, where the input is processed by word-level segmentation and the meaning features are learned. For the second task, a method also based on the BERT model is used. The models trained by the two tasks are used to achieve data mining work in the field of quality problem analysis. Comparative analysis by examples shows that the training results based on BERT model are better than those based on LSTM model in both tasks.