DB2Net: A Deep Learning Approach for Predicting Levels of Interest for Articles Posted on Social Forums
Year: 2024
Type: article
Abstract: Online social forums are crucial for communication and information sharing in today’s digital era. This paper proposes the DB2Net (Dual BERT Decomposed Bilinear Layer Network), a deep learning model designed to forecast the levels of interest (LOIs) for articles posted on Taiwan’s PTT social forum. Using dual BERT modules alongside a newly devised Decomposed Bilinear Layer (DB Layer), DB2Net can explore second-order inter-feature correlations within the textual features of articles. It effectively achieves a prediction accuracy of 98.54% for articles posted within a short-term span of one week, outperforming other BERT models and traditional machine learning models, including XGBoost and Decision Trees. The paper also compares BERT’s performance with that of Bidirectional LSTM, further substantiating the efficacy of using BERT in LOI prediction.
Source: 網際網路技術學刊
Authors: Guan-Wei Chen, Cheng-Chin Chiang
Institution: National Dong Hwa University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4404717657