Personalized Fashion Advice - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 212

•

With the opportunity to shop in front of the computer, two aspects of the shopping
experience are only possible with limitations. The social feedback component has
been almost lost and the chance to try fitting is not existing anymore. We rely on
the description and pictures offered by the shop and comments from other users.
To make sure the ordered item fits, people have to get different sizes of favored
items, making it inevitable that some of the items have to be send back. This is
unpleasant for the customer and expensive for the shop.

In this chapter, we present a use case of an online retailer that aims to improve
the shopping experience of men. By relying on the service of this retailer, customers
no longer have to go shopping in physical stores but can order suitable fashion prod-
ucts online. Differing from conventional online shops where the customers browse
through various products and eventually add items to their shopping basket, this
shopping service relies on the expertise of fashion advisers who, after getting in
contact with the costumers, arrange a combination of different clothes and ship them
to the customers. The customer can then try out the clothes and accessories, keep the
pieces they like and return the unwanted items free of charge.
In order to recommend an outfit, the fashion advisor has to consider the customer's
preferences, context, materials, and a suitable combination of individual items. From
a scientific point of view, we argue that the task of finding matching outfits is far more
complex than that of traditional recommendation systems that recommend movies,
songs or news articles. As outlined in Chap. 5 , many recommendation techniques
rely on the combination of content analysis and collaborative filtering to present
users a list of choices. These techniques rely on computing a group of similar users,
based on the history of purchased or watched items. This approach cannot be easily
applied to the above scenario though since the service aims to take away the burden of
inspecting items from the user. Instead, when recommending clothes, current trends,
the customer's personal style, the occasion for that the clothes are required, as well
as the correct size play an important role. Customers liking the same clothes does not
necessarily mean that they are a good fit or that they are suitable for the user's current
fashion need. Given these constrains, we consider the fashion recommendation task
to be a constraint satisfaction task rather than a collaborative filtering task. The
constraint satisfaction problem (CSP) defines a task where a satisfiable solution has
to be found given a set of constraints. One well-known example for a CSP is the
popular game Sudoku, where the numbers from one to nine have to be placed in a
9-by-9 grid of boxes such that each row, column, and 3-by-3 sub-grid contain each
number exactly once. In our use case, constraints arise as users restrict item choices
according to specific factors such as price, brand, or color. Fashion assembles outfits
which most adequately consider these constraints.
The chapter is structured as follows. We will first introduce the current state-of-
the-art on e-commerce systems with a focus on retailers recommending and sell-
ing clothes in Sect. 8.2 . The related work covers scientific approaches as well as
real-world examples of existing e-commerce applications. In Sect. 8.3 , we introduce
knowledge-based recommender systems, which are the super class of constraint-
based recommendation systems, and explain the basic knowledge required for the