Database Reference
In-Depth Information
Fig. 3.20 Lorenz curve comparing contexts (first row) and lattices (second row) - before (first
column) and after reduction to ranks 8, 5, 3 (remaining columns)
Fig. 3.21 Principle of recommendation experiment
3.3.2 Recommendation and Prediction System
In this experiment, we used the social context of CBD users to suggest or possibly
predict their further actions. Recent paper [ 38 ] shows that it is still an important
problem and the social aspect may be helpful.
Our hypothesis is as follows: the process of dimension reduction in the object-
attribute matrix leads to the unification of objects based on their attributes. Attributes
of an object may change according to the attributes of similar objects. This principle
(illustrated in Fig. 3.21 ) works in the field of latent semantic indexing [ 12 ] and works
also in the field of recommendation systems [ 29 ]. Because users do not behave
randomly, but exist in their own social context and have their own motivation and
knowledge, this principle has a meaningful interpretation here. This hypothesis may
be used to predict user behavior or suggest interesting subjects to the user. We wanted
to analyze the success ratio of such an approach, especially in connection with the
size of the original data used for suggestions.
In order to perform this experiment, we analyzed a random sample of user
accounts of various Wikipedia projects, such as Wiktionary, WikiQuote, Wiki-
Books, WikiSource, WikiSpecies, WikiNews, etc. With each user account, we also
Search WWH ::




Custom Search