International Conference on Computer and Knowledge Engineering

Home / 13th International Conference on Computer and Knowledge Engineering

Dynamic Knowledge Enhanced Neural Fashion Trend Forecasting with Quantile Loss

Authors :

Fatemeh Rooholamini¹ Reza Azmi² Mobina Khademhossein³ Maral Zarvani⁴

1- Department of Computer Engineering, Faculty of Engineering, Alzahra University, Tehran, Iran 2- Department of Computer Engineering, Faculty of Engineering, Alzahra University, Tehran, Iran 3- Department of Computer Engineering, Faculty of Engineering, Alzahra University 4- Department of Computer Engineering, Faculty of Engineering, Alzahra University, Tehran, Iran

Keywords :

Fashion Trend Forecasting،Time Series Forecasting،Time Series

Abstract :

Forecasting fashion trends based on time series and style analysis is an innovative and attractive approach. Still, the dynamic nature of fashion trends and the influence of various factors make this forecast challenging. Additionally, it's important to note that previous research in this field has primarily focused on specific elements such as seasonal clothes, which may not be suitable to the evolving requirements of the fashion industry. So, we extended the Knowledge Enhanced Recurrent Network (KERN) model, which leverages deep recurrent neural networks for time series analysis to forecast fine-grained elements for a specific user group. Due to the low flexibility and the lack of management of sequences with variable memory length in the KERN method and the increasing improvement of the memory efficiency of this model, we propose a Dynamic Knowledge Enhanced Recurrent Network model (DKERN) along with the use of Quantile loss function, which is an improvement on this method. We have tried to help the fashion industry by improving forecasting fashion trends by using various analytical techniques. The proposed approaches have shown improvements ranging between 0.2% to 8.2% compared to previous works in terms of time periods, datasets, and various evaluation metrics.