Fashion Trend Forecasting And Fashion Recommendations Based On Deep Learning

With the rapid development of our economic and material life level significantly increased,the people for their dress is more and more attention.At the same time,the rapid development of Internet technology has prompted a growing number of fashion companies’ main businesses will gradually turn online,more and more people choose online shopping platforms to buy clothes.Therefore,research on online fashion trend forecasting and fashion recommendation methods is of vital importance to fashion companies and fashion consumers.However,due to the explosive growth of online fashion information,and the update cycle of fashion clothing is becoming shorter and shorter,fashion companies need to obtain fashion element trends from previous data in time,and fashion consumers need to analyze the clothing they need from massive data.Based on the remarkable success of Deep Learning technology in various fields,this paper takes the specific requirements of clothing companies and fashion consumers as the starting point to do the following research work:First,this paper is based on the fact that traditional fashion trend forecasting is highly dependent on the subjective will of experts and users,and that major clothing brands will release the next quarter’s clothing one quarter in advance in the four famous world-class fashion weeks held every year,to guide the fashion in the next year.The current state of fashion trends in the first quarter.This paper proposes a time series-based clothing fashion trend forecasting method.The method firstly crawls the four fashion week show pictures from 2013 to 2018 in the fashion website vogue and extracts three fashion properties from the images through the commercial fashion label website visenze.com.Various types of fashion features（categories,attributes,and styles）are used to construct a fashion week dataset,which contains a total of 152,750 fashion week show images and their corresponding fashion elements.Then analyze the fashion week data set to extract the knowledge of the fashion week show;use the Manhattan distance to measure the time series distance between each fashion element to obtain the internal knowledge of the clothing;finally,the Seq2 Seq model based on the Attention mechanism will extract the fashion week show picture information.At the same time,the internal knowledge of fashion is combined to find fashion relationships from time series,to predict fashion trends.When this method predicts the trend of fashion elements in the second half of the year,in the Geo Style dataset,MAE reaches 0.130,MAPE reaches 14.12;in the FIT dataset,MAE reaches 0.082,MAPE reaches 29.42,and its experimental results are better than other models.Secondly,this paper proposes a generative fashion recommendation method based on multi-modal feature fusion（Multi-Modal Fashion Recommendation Network,MFR）.The method consists of four modules: semantic extraction,expert system,feature fusion,and fashion generation.First,the semantic extraction module converts the visual information in the input existing clothing images into semantic information;then,it analyzes the existing fashion matching data to construct an expert system with matching rules for fashion elements;finally,the expert advice and User requirements are combined with features to obtain optimal recommendations,and integrated with visual information of clothing into fashion generation,providing users and designers with an excellent fashion recommendation solution.The experimental results show that this method can recommend and generate clothing fashion items that balance the compatibility between users’ current requirements and fashion items,and the AUC can reach 89.2%.Finally,fashion elements play an important role in both fashion trend forecasting and fashion recommendation.Excellent fashion trend forecasting can also affect the description of fashion consumers’ requirements in fashion recommendation and promote the personalization of fashion recommendation balanced with fashion.In real life,it helps to increase the sales of fashion companies and the online shopping experience of fashion consumers,thereby promoting a virtuous circle in the fashion industry.