The Big Picture: Toward a Synthesis of RL and Adaptive Tensor Factorization - Realtime Data Mining - page 206

Chapter 10
The Big Picture: Toward a Synthesis of RL
and Adaptive Tensor Factorization
Abstract We explore the subject of uniting the control-theoretic with the
factorization-based approach to recommendation, arguing that tensor factorization
may be employed to vanquish combinatorial complexity impediments related to
more sophisticated MDP models that take a history of previous states rather than
one single state into account. Specifically, we introduce a tensor representation
of transition probabilities of Markov-k-processes and devise a Tucker-based
approximation architecture that relies crucially on the notion of an aggregation
basis described in Chap. 6 . As our method requires a partitioning of the set of state
transition histories, we are left with the challenge of how to determine a suitable
partitioning, for which we propose a genetic algorithm.
In this research-oriented chapter, we shall study a refinement of the concept of a
Markov decision process which enables a recommendation engine to incorporate
sequences of previously visited products rather than making decision exclusively
upon the current state. As foreshadowed in Chap. 8 , more sophisticated models of
this kind entail some complexity issues. Therefore, so as to vanquish the latter,
we shall introduce a tensor factorization-based approximation framework. The
reasoning provided in this chapter is thus a step toward a unification of classical
(factorization based) data mining on one hand and the novel control-theoretic
framework on the other hand. We should stress, however, that the approach
presented in the following is currently still in its infancy and a subject of ongoing
research. Hence, a major part of the subsequent elaborations are still based upon
speculation rather than scientific rigor.