Information Technology Reference
In-Depth Information
vector machines (SVM). The results of experiments demonstrate that the iteractive
pattern tree kernel method is effective in extracting PPI. In addition, the proposed
interaction pattern generation approach successfully exploits the interaction semantics
of text by capturing frequent PPI patterns. Consequently, the method outperforms the
tree kernel-based PPI method [3, 9, 13, 17]; the feature-based PPI method [1, 16]; and
the shortest path-enclosed tree (SPT) detection method which is widely used to
identify relations between named entities.
2
Our System Architecture
Figure 1 shows the proposed interaction extraction method, which is comprised of two
key components: interaction pattern generation and interaction pattern tree
construction . We regard interaction extraction as a classification problem. The
interaction pattern generation component aims to automatically generate representative
patterns of mention interactions between proteins. Then, the interaction pattern tree
construction integrates the syntactic and content information with generated interaction
patterns for representation of text. Finally, the convolution tree kernel measures
similarity between interaction pattern tree structures for SVM to classify interactive
expressions. We discuss each component in detail in the following sections.
Fig. 1. The interaction extraction method
3
Interaction Pattern Generation
The human perception of a protein-protein interaction is obtained through the
recognition of important events or semantic contents to rapidly narrow down the
scope of possible candidates. For example, when an expression contains strongly
correlated words like " beta-catenin ", " alpha-catenin 57-264 " and " binding "
simultaneously, it is natural to conclude that this is a protein-protein interactive
expression, with a less likelihood of a non-interactive one. This phenomenon can
explain how humans can skim through an article to quickly capture the interactive
expression. In light of this rationale, we proposed an interaction pattern generation
Search WWH ::




Custom Search