Year: 2024
Type: conference-paper
Abstract: In the Internet era, video has emerged as a novel form of media, with its commercial value directly linked to the attention it receives. For media creators, predicting a video's popularity can help clarify creative direction. For video platforms, accurately estimating a video's popularity enables the efficient allocation of promotional resources. Predicting video popularity assists these groups or teams in achieving greater economic returns. To enhance the flexibility and explainability of predicting video popularity, this study proposes a video text feature explainability popularity prediction method based on RoBERTa (R-VEPP). This study collected publicly available information from the Bilibili website and used the bibiGPT video summary interface and the Ernie large language model to create a self-constructed text datasets. The author developed a prediction model by leveraging RoBERTa and cosine similarity calculations. Ultimately, the model demonstrated outstanding performance in the automatic scoring task, with a mean squared error (MSE) of 0.5728 and an accuracy rate of 88.672%. R-VEPP enhances the flexibility of video popularity prediction while achieving a certain degree of explainability in scoring. The results indicate that R-VEPP holds practical value.
Author: Jingwei Zhang
Institution: Southwest Jiaotong University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4405849295