There are two basic types of nodes: learning (or statistical) nodes and rule nodes.
Learning nodes collect statistics and learn from examples. Learning nodes learn from the feedback they receive and return scores based on this feedback. When IBM® Content Classification receives a new text for classification, it will return higher scores for nodes that received feedback on similar texts.
Positive feedback to a node has a positive effect on the node, since it tends to increase the score that this node will return when similar texts are presented in the future. Conversely, this positive feedback will have a negative effect on all other relevant learning nodes that can return scores.
A principal learning node is used to create a statistical hierarchy between sub-nodes. Most learning nodes are not principal nodes, regardless of their location in the tree. By marking a node as principal, you can create an internal statistical hierarchy that affects the relationship between statistical nodes. For example, in some cases placing two categories that are "stealing" from each other under a principal learning node might improve performance. For more information, contact IBM Software Support.
Rule nodes use fixed rules to return a score. For example, the rule node "If language = English" will return a score of "True" when the language is English. Rule nodes do not learn from feedback.
Learning nodes and rule nodes cannot be direct siblings (that is, they cannot occupy the same level in the tree). The Knowledge Base Editor will not allow you to move a rule node under a node that already contains a learning node.
Each type of node can have either one or more rule nodes OR one or more learning nodes beneath it. A rule node with one or more learning nodes beneath it becomes a principal rule node. Likewise, a learning node with one or more rule nodes beneath it becomes a principal learning node.