Recommender Systems, Mathematics and Graphs
Time: Wed 2014-05-14 13.15 - 14.00
Lecturer: Roelof Pieters, Vionlabs R&D
Location: in 4523, Lindstedsvägen 5
Recommender Systems have become extremely common in recent years. A whole
subfield of science, a yearly conference, and widespread use in industry,
has pushed forward its development. This short seminar will focus on the
various uses of Recommender Systems (RecSys), and the different approaches
towards information selection, generally understood as Collaborative
Filtering, Content-based Filtering or Context Filtering.
The talk will give insight in some of the mainstream mathematical models
used to solve various problems in information selection and representation,
ie. cold start, scalability, sparsity, similarity versus diversity, as well
as some more experimental ones developed at Vionlabs.
After a general introduction into the main principles and algorithms used in
RecSys in its various shapes, it might if time allows also show how graph
theory and graph databases have leveled the playing field, making it
possible for low budget machines to sift through and do calculations on
terabytes of data, before only privileged to large-scale enterprise-type
machines. Here the talk will also point to some of the faster approximation
algorithms possible by modeling data as graphs.
As the talk is part of admittance to the PhD program of Theoretical Computer
Science, it will be heavy in math, while still trying to explaining the more
general dynamics and uses of the formulas and algorithms sketched. In some
cases IPython Notebook will be used to show actual working code of typical
RecSys algorithms (Clustering, Matrix Factorization, etc.) on a sub-set of
the MovieLens dataset.
Roelof Pieters is an anthropology-trained language and computer nerd,
currently working on a next-generation recommendation and search engine –
FoorSee - at Vionlabs (www.vionlabs.com) in Stockholm, and as such heads its
R&D department, and is now excitedly hoping to pursue his PhD in Theoretical
Computer Science @ KTH.