Labels are nodes whose activation value modifies the influence of a node on another one.

```                  +------------+
| opposite (3|
+------------+
+---------+              |               +-----------+
| happy (1|<-------------+-------------->| unhappy (2|
+---------+                              +-----------+
```

Let C1 be the occurrence number of the word 1 -happy-
Let C1,2 be the co-occurrences number of the words 1 and 2.
Let A3 be the activation level of word 3 -opposite, the label node-
Let I1,2 the percentage of influence of the node 1 on the node 2.

I1,2 = C1,2 / C1 + (1 - C1,2 / C1) x A3

In this example, there is a link between happy and unhappy, that are closely related (in real world), but not conceptually similar.

This facts could be translated in:

not conceptually similar
Their number of co-occurrences is low, related to their number of occurrences (C1,2 < C1 and C1,2 < C2)
closely related
when you look for the opposite of happy, you should find unhappy: when the node happy is activated (Ahappy = 100) and the node opposite is activated (A3 = 100), then, after activation propagation, using the formula for influence given upper, the node unhappy will be activated.

For example: say that happy occurred 16 times in the corpus used to build the Concept Network, that unhappy occurred 3 times in the same corpus, and that they appeared 2 times in the same sentences, here are the value to give to the variables:

• C1 = 16
• C2 = 3
• C1,2 = 2
Let Ahappy and Aopposite be 100%, then

I1,2 = 2 / 16 + (1 - 2 / 16) x 100% = 100%

On the contrary, when AVopposite = 0%, I1,2 = 2 / 16 + (1 - 2 / 16) x 0% = 12.5%, so that when happy is activated, unhappy is not much activated through activation propagation.

Log in or register to write something here or to contact authors.