Method of tuning a decision network and a decision tree model

A method of tuning a decision network comprising:

• in a decision network comprising a hierarchical set of nodes and leaves, identifying a path of nodes traversed through said decision network resulting in an interpretation of a data item, wherein said nodes correspond to existing queries;
• calculating a measure of goodness at a node in said path using a new query as a replacement for one of said existing queries;
• and
• if said new query results in an improved measure of goodness compared to said existing query, retraining said decision network using said new query at said node beginning with said node through said leaves of said decision network.

A machine-readable storage, having stored thereon a computer program having a plurality of code sections executable by a machine for causing the machine to perform the steps of:

• identifying a deviation between a correct interpretation of a data item and an incorrect interpretation of said data item;
• in a decision network comprising a hierarchical set of nodes and leaves, determining a path of traversed nodes in said decision network resulting in said deviation, wherein said nodes correspond to queries;
• calculating a measure of goodness at a node in said path using at least one new query as a replacement for an existing query to determine whether said measure of goodness improves using said at least one new query compared to said old query;
• and
• if said measure of goodness improves, selecting said at least one new query and regrowing said decision network from said node down through said leaves using said at least one new query at said node.

A machine-readable storage, having stored thereon a computer program having a plurality of code sections executable by a machine for causing the machine to perform the steps of:

• in a decision network comprising a hierarchical set of nodes and leaves, identifying a path of nodes traversed through said decision network resulting in an interpretation of a data item, wherein said nodes correspond to existing queries;
• calculating a measure of goodness at a node in said path using a new query as a replacement for one of said existing queries;
• and
• if said new query results in an improved measure of goodness compared to said existing query, retraining said decision network using said new query at said node beginning with said node through said leaves of said decision network.