Fashion Trend Forecasting Based on Multivariate Attention Fusion

The objective of garment fashion trend prediction is to capture the trends of different garment attributes such as round necks and camouflage, enabling forecasts of their future popularity. Existing fashion trend prediction models have not sufficiently integrated comments on current social media networks and user preferences. Thus affecting the accuracy of garment popularity prediction. To address this issue, this paper proposes a fashion popularity prediction model based on multivariate attention fusion(MAFT). It combines diverse information posted by users on fashion platforms like Chictopia, uses GLU modules and dilated convolutions to preprocess multivariate features, enhances context feature extraction on sequence data, and suppresses irrelevant information. Subsequently, a multivariate attention fusion block is designed to capture the mapping relationship between dynamic and static variables in the input. After feature fusion, trend prediction for the garment is achieved through a Transformer layer. Experimental results demonstrate that this method accurately predicts future trends on the SFS, FIT, and Geo datasets, with improvements of 8.79% and 11.77% in MAE and MAPE evaluation metrics, respectively, compared to the best existing fashion trends prediction models.