A Study on News Headline Classification Based on BERT Modeling
Yucheng Chen · Advances in computer science research · 2024
News is an important way to understand the information of contemporary society, and it is necessary to quickly categorize and identify a large amount of news information.In this report, a classification task was performed on Chinese news headlines based on the Bidirectional Encoder Representations from Transformers (BERT) model.Deep learning model transformers are used to compare the differences between Bert model and traditional methods in text categorization.Training and tuning were performed on the collected and organized dataset.From the experimental results, the model has a better classification effect on news headline classification, reflecting the advantages and performance of Bert in Chinese news headline text classification.Meanwhile, the performance differences of Bert model under different learning rate parameters, number of learning rounds and different dataset annotation accuracy settings are analyzed.The results of the experiment were 0.7 accuracy for 10 rounds of learning with a learning rate parameter of 5e-6, and 0.6 accuracy for 20 rounds of learning with a learning rate parameter of 1e-5.The analysis concludes that under the same learning rate parameter, the learning accuracy tends to stabilize with the increase in the number of learning rounds; under the same number of learning rounds, the learning rate is too high or too low will affect the learning accuracy.