This thesis describes what recommendation systems are, what they are used
for, how to build one, and what the potential business models around them are.
The work starts with a literature review of recommendation system usage,
recommendation system algorithms, and how to evaluate them. As empirical
part of the thesis, implementation effort of a prototype recommendation
system is described. The prototype is then evaluated in terms of
scalability and quality of recommendations.
Business part of the thesis discusses business models that are based on
focusing around different parts of a typical e-commerce process which
employs recommendation systems. The models are examined in two dimensions:
customer segment type (business or consumer), and for company of what size
they are applicable for.
The main conclusion of this thesis is that recommendation systems seem an
important part of business. They also seem not very hard to implement,
although selling one seems challenging based on failure to sell the
produced prototype. It seems puzzling that there does not seem to be much
business or even academic literature along the lines of business models
noted within the thesis.
In technical implementation, there are obviously some pitfalls, and as the
prototype never reached product quality, they may be perhaps
under-emphasized. Some interesting work was produced on how to actually
design and implement a highly scalable hybrid recommendation algorithm,
with linear computational complexity scaling based on number of users,
items and rankings to be processed.
