Bert-based model for Ceramic Emotional Text Recognition and Multi-label Classification
Busheng Li, Siqi Li, Wangren Qiu, Tao Liu · Jisuanji shenghuojia. · 2024
Ceramic products are necessities for daily home life, while the emotional analysis research on ceramic products cannot keep up with the needs of social and economic development. To mine the valuable information on ceramic emotion quickly and effectively, this work has proposed a novel method for text recognition and classification of microblogging information related to Jingdezhen ceramics by fusing Bert model and multi-label classifiers on the novel dataset we first established. Firstly, the first multi-label emotion dataset consisting of 7564 samples was constructed on 10154 raw samples obtained with Python from microblogging information, which is noted as the keyword “Jingdezhen ceramics”. Secondly, the data were categorized with multiple labels into useful or useless information. Useful information was further classified into three categories: price information, appearance information, and quality information. Thirdly, a hybrid model combining Bert and multi-label classifiers has been provided for analyzing ceramic emotion. Based on a series of experiments, it has been observed that the predictor’s accuracy is 92% and its loss value is 0.33. This work highlights the annotation guidelines, dataset characteristics, and insights into different methodologies used for analyzing ceramic emotion and cultural heritage.