Information Technology Reference
In-Depth Information
of node 4, as node 4 is part of the explanans elicited by the why-q; thus, general
information contained in nodes 2 and 3 is ruled out. Only nodes 5 and 6 remain.
In short, if we take the graph of the figure 17.6 and launch it against the query
Why John died?
some answers arise. The graph has a main node or prior cause
that directly or indirectly connects with the effect. So, a tentative answer could be
to locate the root node and say:
Because John was a smoker.
But, as previously
said, why-questions are concerned - if accessible - with deepening in knowledge.
Thus, other nodes should be explored. Our hypothesis is that central nodes in the
mechanism include relevant content. In the quoted graph, the central node is node
4. But we can tune perhaps a little more. A why-question can be made by a skilled
or a non-skilled, interrogator. If the questioner is not specialized, the causal link be-
tween the prior cause and the central node is perhaps enough to arrange the answer.
As In the previous example,
Because John was a smoker, causing lung cancer
.But
if the questioner is specialized, more specific knowledge is needed. Eingenvector
centrality values detect nodes connected with a few neighbors of high importance.
If the node with the highest centrality value is node 4, the nodes with more eingen-
centrality are 1, 2, 5, 3 and 6. Node 1 is ruled out by the root cause. Nodes 2 and
3 do not include specific content, as they are before node 4, the central node. Thus,
nodes 5 and 6 are the candidates to express specific explanations about why lung
cancer causes death. Thus, the answer for a specialized questioner will be the fol-
lowing causal chain:
Because John was a smoker, provoking lung cancer, that leads
to fluid collect or bleeding
(
leads to a
synonym of causing, and included in the final
answer in order to make it more linguistically expressible).
To provide an automatic answer, we have modified partially the algorithm used
with how-questions. This procedure selected the possible paths linking the node
cause with the node effect. Each one of these paths would be a cluster of nodes
to summarize.
So, the steps to answer (and summarize) a why question are the
following:
•
Locate the main cause node (head of the diagram), and the effect node.
•
Calculate the centrality measure of each node (but the cause and effect nodes).
•
Select the node with the highest centrality value.
•
Select those nodes with the highest eingencentrality value related to the node
with the highest centrality value.
•
Reject those nodes selected in the previous step which number is lower than the
node with the highest centrality value.
•
Order the nodes to compose an appropriate answer.
•
Compose an answer summarizing the retrieved nodes.
In the graph showed in figure 17.6, the algorithm would locate nodes 1 and 9 as
cause and effect nodes. The node with the highest centrality value would be node 4,
so the eingencentrality values will be calculated in base to this node. As a result four
more nodes are obtained in this order, 2, 5, 3 and 6, but nodes 2 and 3 are rejected
because they are lower than 4. On the other hand, nodes 5 and 6 will be included
in the new answer, as well as the causal paths that link these nodes with the effect
node.
