Albert Based Policy Topic Classification Model
Min Tan, Yang Cao, Ziqi Xiong, Dezhi Kong, Yankai Xue · 2023
As an important government information resource, polices are the sum of rules and actions customized by the governments, political parties or other organizations to achieve their goals. In today’s explosive growth of data, in order to quickly retrieve policies related to established topics from massive data, we proposed a policy topic classification model based on the pretrained language model Albert. The model first pre-trained on an unsupervised large corpus, and then fine-tuned in a small amount of labeled data to improve the classification effect of policy topics. The experimental results show that this method can achieve automatic classification of policy topics, with classification results of macro average precision 92.74%, macro average recall 91.68%, and macro average F-Score 92.21%.