Building a Recommendation Engine: The XELOPES Library - Realtime Data Mining - page 294

selRec ¼ nextSelRec;
nextState ¼ cstate;
nextRecs ¼ crecs;

}

12.3 Application Example of XELOPES: The prudsys RDE

We conclude this chapter by a practical example for the application of the
XELOPES library in recommendation engines. The prudsys Realtime
Decisioning Engine (RDE) is a realtime analytics system developed by prudsys
AG, which follows the principles of realtime analytics described in this topic. Its
range of functions extends well beyond that of a standard recommendation
engine.
It comprises the following six modules:

1. RDE | Recommendations : realtime recommendation engine
2. RDE | Newsletter : realtime newsletter personalization
3. RDE | Pricing : realtime price optimization
4. RDE | Assortment Planning : realtime planning
5. RDE | Scoring : realtime scoring
6. RDE | Search : realtime search

This topic concentrates primarily on the way in which the RDE | Recommen-
dations module works, specifically on the adaptive learning, and - first of all - on
the reinforcement learning aspect. However, the module also includes many
other algorithms, such as basket and sequence analysis, collaborative filtering,
item-to-item collaborative filtering, singular value decomposition, and tensor
factorization.
It is interesting to note that the concept of realtime learning has now been applied
to many new functions across the other modules (so the “realtime” prefix is more
than just a marketing ploy). For example, the RDE | Scoring module allows realtime
scoring as described in Chap. 7 . The RDE | Assortment Planning module updates its
planning model continuously throughout the day, allowing delivery to be brought
forward by a day, for example. Probably the most impressive module, RDE |
Pricing , which carries out dynamic price optimization, is one of the key inventions
of the prudsys AG. It varies retail product prices in real time (or at periodic
intervals) according to user behavior and demand, with the aim of maximizing
profits. Reinforcement learning is used here, too. For more details, see the separate
pricing module white paper [Lip11].
The switch to real time not only offers an entirely new standard of quality for
analytics (or actions) but also opens up entirely new business scenarios. Welcome
to the RDE realtime world!
The layer model on which the prudsys RDE is based is shown in Fig. 12.18 .