Overview and details of the sessions of this conference. Please select a date or room to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).
This agenda is preliminary and subject to change.
Please note that all times are shown in the time zone of the conference. The current conference time is: 19th May 2024, 10:54:32pm CST
Session Chair: Rong ZHENG, Jilin University Session Chair: Jiuming Ji, East China University of Science and Technology
Location:Room 4
Events III on 3F
3F沙龙III
Presentations
GSBERT:An Empirical Study of an Automatic Quality Detection Method for Datasets Based on Visual Interpretation
T. Zhang, T. Yu, H. Ma, L. Jiang
Heilongjiang University, China,
The development of a large-scale pre-trained model has opened a new stage for artificial intelligence. More and more researchers have introduced models such as BERT and GPT into various application fields. However, the problems of complexity and interpretability have posed a degree of risk to the model’s generalization. Meanwhile, the accuracy of the model directly depends on the quality of the training dataset. Optimizing the dataset often leads to better results. Therefore, this article proposes the GSBERT model. Firstly, the model can explore a decision basis of the BERT model by gradient saliency. It can also present the BERT model decision-making process by visualization, so as to makes model decisions interpretable. Secondly, based on the interpretability of the model, the sample feature token is extracted to form a feature word list. Then classify and predict samples by Naive Bayes model again. Lastly, to test the effectiveness of GSBERT in analyzing data consistency, the GSBERT model is used to automatically detect and re-label the misclassified samples of the news classification datasets. The accuracy of model judgment also improved by 12.55% on the validation set due to the improved quality of the dataset. Not only can this study guide dataset quality detection in pre-trained models, but also can improves the accuracy of the model by peering at each layer during the running status of the BERT model. This method aims to provide reference ideas for the wider application of deep learning models in various fields.
网络圈层化背景下社交媒体用户信息茧房滞留行为研究
J. Dong1, F. Bai2, D. Wu2
1Central China Normal University, China, People's Republic of; 2Wuhan University, China, People's Republic of