The Collaborative Fashion Recommendation System

Posted on : by : admin

The Collaborative Fashion Recommendation System

The Collaborative Fashion Recommendation System (CFRS) we have implemented focuses exclusively on clothing items. It combines features that are found in online fashion stores and other recommendation systems (metafeature). First, like any online store, users can browse through various product categories. For each clothing item the following features and product details are represented: One or more photos of the product, material, color, brand, price, description and trend score, that defines how trendy the product is now. Aleesha Institute Fashion Designing The calculation of trend scores is presented in the subsequent sections of the paper. Moreover, users can get ideas about how they may combine a piece of clothing with other clothing items from other categories. Our main goal is helping the user get ideas and making a complete and trendiest outfit. System architecture The CFRS system architecture is depicted in figure 1. It consists of a database management subsystem (written in MySQL), while the web front-end consists of three subsystems: • Products management subsystem • Products presentation subsystem • Trend score calculation subsystem. The following user categories are supported by CFRS:

Administrator: The System administrator is responsible for products management (add, edit, delete, update functions available), trends management and users management. To add a product, admin fills-in details such as color, material, price, description, photos etc. Besides products management, the administrator is responsible for managing trends. Trends are related to current year, divided into two semesters: Autumn/ Winter and Spring/ Summer. This distinction was based on the fact that clothing presentation in fashion weeks and therefore, trends definition is related to these time intervals. The administrator has the ability to add trends for the following three categories: Colors, prints and materials. After selecting the desired trend category, she/he chooses one of the values and the season that corresponds to this trend. Also, the administrator can edit or delete the available trends.

Fashion experts: In recommendation systems and especially in those used in e-shops, fashion trends are particularly important, but clothing items play a primary role. Likewise, in CFRS fashion trends and classic clothing item parameters are equally important. Having this in mind, the second user category in our recommendation system are fashion experts. This category consists of fashion magazine editors, fashion designers, fashion bloggers etc. Fashion experts register in the system as simple users and then an administrator assigns this role to them. Once logged in, experts can rate fashion trends. To rate them, they select one of the three categories, which are defined by the administrator so only the trends of this specific category are displayed. Fashion experts vote whether they like or not each trend by clicking the like or dislike button. Experts can rate trends and change their vote whenever they want. Nevertheless, they can rate each trend just once. An important feature is the ability given to experts to observe how other rated trends evolve while keeping some rating information hidden (for example individual expert rating for each trend is not made public).

Users: As soon as simple users log in the recommendation system, they can be informed about current fashion trends. Also, they can, follow (like) experts, and they can also rate fashion trends by following the same procedure as experts.

Visitors: Visitors can browse the different categories of clothing and use the recommendation system. Also, contrary to domain experts and users, visitors can’t see or rate fashion trends. A. The rating mechanism As mentioned previously, our recommendation method depends on each product’s trend score, as calculated by ratings in the system (fashion experts and users).

is added to the system. A trend score is calculated when the administrator adds a clothing item, taking into account the following attributes: Color, material and print. These attributes are selected because they refer to trend categories contained in the recommendation system. Afterwards, the values of these attributes are compared with trend values (figure 2). In case that a trend value coincides with any of material, color or print, the trend score for this value is calculated by adding the individual votes. The score coming from votes of experts is calculated separately from that of simple users. Then they are added. In this way the score for each attribute is calculated. It’s worth mentioning that votes coming from experts have a 50% more weight compared to votes of simple users. Eventually, the overall trend score for each product is the sum of individual color, material and pattern ones.

Product details: Besides from product details such as price, material, brand etc. system’s users or visitors can view the trend score of each product.

Sorting products: Our recommendation system gives the opportunity to those who use it to sort the product list according to how fashionable these clothes are, based on trend scores. So, users can see the trendiest and least trendy (or more classic piece) in every clothing category according to current trends. In addition, it is possible to sort them by the classic price-based option.

Recommendations based on trend score: As a recommendation system, CFRS provides outfit suggestions for all products contained in the system. All suggestions are based entirely on the trend score. The recommendations made, combine the clothing item the user is interested in, with other products of different categories. For example, assume that a user is interested in a specific blouse. By using our system, she can get recommendations for skirts, trousers, jeans, coats and jackets that can be combined with this blouse (figure 3). The suggestions are made depending on how fashionable the products are, based on current fashion trends, using the trend score. Recommended products have the highest trend score in their category.