Automatic Extraction of HLA-Disease Interaction Information from Biomedical Literature - Advances in Computational Science and Engineering - page 225

3. Associated is recognized as relation keyword and it is copied to all of its
parent nodes. (associated → VBN → VP → S1 → SBAR → VP → VP →

S2)
4. HLA-DQ2, DQA1*0501 and DQB1*0201 are recognized as HLA entities
and they are copied to all of their parents. (HLA-DQ2, DQA1*0501 and
DQB1*0201

VP →

→ NPB → PP → VP → S1 → SBAR → VP → VP → VP →

S2 )
5. When searching algorithm reaches the right end of terminal nodes
(DQB1*0201), it goes up to nonterminal nodes from right to left. (NPB
→

S1 )
6. When searching algorithm visits S1, the first nonterminal node with HaD
information, it returns the interaction information (Celiac disease associ-
ated HLA-DQ2, DQA1*0501 and DQB1*0201) and the whole procedure is
terminated.

PP VP VP

→ → →

Unlike HaD words, filtering keywords play a different role in the information
extracting procedure. Filtering keywords keep the searching algorithm from ex-
tracting information from little informative sentences like the immune response
test. As filtering keywords are common words and have meaning only when they
appear near HLA entities, the algorithm checked where filtering keywords ap-
pears in a parse tree. If filtering keywords appear closer to HLA entities than

Fig. 3. Simple information extraction process