Abstract:

The availability of high dimensional data has increased significantly
over the last few years. To represent this kind of data graphs are
frequently used, due to its additional expressiveness and storage
efficiency. Semantic data is often stored in such structures, since
multiple dimensions are often necessary to capture the various aspects
of knowledge. Depending on the nature of the high dimensional data, the
distribution of relations among elements can capture a large portion of
the semantics of those elements. However, knowledge discovery based on
analyzing patterns of relations is rarely used. In that context we
propose various inference based algorithms exploiting the semantic
properties of hierarchically represented knowledge. We implement such
algorithms with the goal of discovering new elements and relations in a
hierarchical graph. We test those algorithms on a generalized version of
the Cyc's knowledge base and discuss the significance of the resultant
semantic learning. Finally, we argue why such algorithms can be useful
for unsupervised learning, and supervised analysis of knowledge bases.