With the popularity of short videos, “short video e-commerce” are adopted by more and more companies. Short videos are very helpful in e-commerce context because they carry more information than texts and pictures do. However, what kind of short videos can more effectively promote product sales in e-commerce context? This paper draw samples o£ short videos on the "Mogujie" platform and extracts features that may affect the marketing effects of short videos, such as visual perception, auditory perception and video content. Then we construct multiple linear regression model, classification and regression tree model and random forest model to study the relationship between short video features and their marketing effects, which are measured by the sales of the products introduced in the short video. The research results show that although the visual, auditory and content features of short videos are all related to the marketing effects, the content features have the strongest effects on the sales performance.