1.4
Contributions
9
Methods for User Representation
User representation methods proposed are agent-based abstractions of users to-
gether with their corresponding profiles (paper A, B and C), usage logs for
MMOG services (paper E) and related empirical results (paper D).
Methods for User Prediction
User prediction methods proposed are classifier algorithms (paper F, G and H)
and related empirical results on user prediction with user representation data
in the form of usage logs from an actual web site and the simulated MMOG
setting from paper C (paper I). Note that the classifier algorithms proposed are
not limited to user prediction, but also as general incremental linear classifiers.
The main contributions of this thesis are:
C1:
A conceptual solution and a supporting platform for implementing and
using personal software assistant agents in mobile commerce services. The
solution is aimed towards relaxing the restrictions of mobile devices and
wireless communications (Paper A).
C2:
A conceptual peer-to-peer extension of the platform for supporting scalable
and distributed product and service recommendations for mobile commerce
customers (Paper B).
C3:
A scalable platform for simulating customer user in particular kind of
mobile/electronic-commerce service - Massively Multiplayer Online Games
(Platform described in paper C and the related empirical performance eval-
uation in paper D).
C4:
Investigation and proposition of requirements for doing customer person-
alization in Massively Multiplayer Online Game services. This includes the
creation of a definition of data mining types for MMOGs, a classification
of computer games from a data mining viewpoint, a comparison of infor-
mation gathering for web usage mining and game usage mining, and finally
a proposal for a common game log format to enable game usage mining
(Paper E).
C5:
Investigation and proposition of algorithms that can be used in m/e-
commerce personalization, including developing classification algorithms
that: scale with a large number of classes (Paper F), utilizes parallelization
(Paper H), and handles changes in classification data over time (Paper G).