An Explainable XGBoost-based Approach for Assessing Product Presentation Video

Abstract

Product presentation video, an essential medium for online product presentation, mainly conveys product information to consumers. The quality of product information expression will affect consumer's purchase experience, even purchase decisions. But many video production novices often have difficulty assessing the effectiveness of their videos and doing optimization. Thus, we adopt information mediums' perceived usefulness and perceived ease of use as assessment indicators, and propose an automatic explainable assessment method. The method is based on the combined use of the extreme gradient boosting algorithm and SHapley Additive exPlanations method, which is implemented to assess video samples and generate explanations on the assessment results. We create a product presentation video dataset for development and evaluation. The results show that our method can assess video effectively, and provide insights into the shortcomings of videos' production like experts.