Personalized Fashion Advice - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 215

•

The context of the user request for clothes: For example, leisure or business clothes.
Preferences for clothes can be learned for different contexts.

• The fitting of clothes: Even if a cloth matching the user's preferences, it is still

important that the clothes fit.
With this knowledge in mind, we propose our approach in Sect. 8.5 which uses
a constraint-based recommender to incorporate and satisfy the above-mentioned
aspects. Before we detail our approach, we first introduce the concept of constraint-
based recommendations, which are a subset of knowledge-based recommendation
systems, in the next section.

8.3 An Introduction to Knowledge-Based
Recommender Systems

Knowledge-based Recommender Systems are another major category of recom-
mender systems besides collaborative and content-based recommender systems [ 9 ].
Knowledge-based recommender systems are typically used in domains where the
items to recommend are not bought or interacted with very often. Movies or books
for instance are domains where a lot of people buy or rate them and thus, information
as input for content or collaborative approaches is given. In domains where items
are bought less frequently, such approaches are not the best suited ones. The given
scenario, an online shop for clothes specialized on men, is such an scenario, where
only few information is given about the user themself, see Sect. 8.4.1 .
Knowledge-based RS combine knowledge about itemattributes, domains and user
preferences. They seek to determine whether a particular item suits the user's needs.
For instance, a vacation location where it is warm and affordable [ 18 , 20 ]. One can
distinguish between two types of knowledge-based recommenders—case-based and
constraint-based recommender. Constraint-based RS try to find items that exactly
matches users' requirements using a predefined knowledge base containing rules
how to relate user requirements and items. Case-based RS on the other hand utilize
similarity metrics to match user preferences with item descriptions.
In this work, we employ case-based recommender to compute recommendations.
Case-based recommender (CBR) utilize similarity functions to find a set of items
matching the users' queries, needs and/or preferences. CBR rely on a set of known
cases, the case base. CBR use the case base to adapt or transfer the knowledge from
previous cases to find selections of items satisfying the current recommendation
request [ 7 , 30 ]. Often, user requests for an item with certain attributes cannot be
successfully handled. Either because the request of the user contains conflicting
requirements or there is no matching item in the item base.
Current research works on methods for intelligent relaxations of constraints or
repairing inconsistencies [ 19 , 26 ]. An early application using a relaxation approach
is OpAmps , which helps customers finding amplifiers, not by exact matches between
the customer preferences (often those preferences are to specialized and cannot be
fulfilled), but by finding the best matching products [ 47 , 48 ]. One of the first examples