RULE_CLASSIFIER

Use the RULE_CLASSIFIER type for creating preferences for the query rule generating procedure, CTX_CLS.TRAIN and for CTXRULE creation. The rules generated with this type are essentially query strings and can be easily examined. The queries generated by this classifier can use the AND, NOT, or ABOUT operators. The WITHIN operator is supported for queries on field sections only. Table 2-41 lists the attributes for the RULE_CLASSIFIER type.

Table 41 RULE_CLASSIFIER Attributes

Attribute	Data Type	Default	Min Value	Max Value	Description
THRESHOLD	I	50	1	99	Specify threshold (in percentage) for rule generation. One rule is output only when its confidence level is larger than threshold.
MAX_TERMS	I	100	20	2000	For each class, a list of relevant terms is selected to form rules. Specify the maximum number of terms that can be selected for each class.
MEMORY_SIZE	I	500	10	4000	Specify memory usage for training in MB. Larger values improve performance.
NT_THRESHOLD	F	0.001	0	0.90	Specify a threshold for term selection. There are two thresholds guiding two steps in selecting relevant terms. This threshold controls the behavior of the first step. At this step, terms are selected as candidate terms for the further consideration in the second step. The term is chosen when the ratio of the occurrence frequency over the number of documents in the training set is larger than this threshold.
TERM_THRESHOLD	I	10	0	100	Specify a threshold as a percentage for term selection. This threshold controls the second step term selection. Each candidate term has a numerical quantity calculated to imply its correlation with a given class. The candidate term will be selected for this class only when the ratio of its quantity value over the maximum value for all candidate terms in the class is larger than this threshold.
PRUNE_LEVEL	I	75	0	100	Specify how much to prune a built decision tree for better coverage. Higher values mean more aggressive pruning and the generated rules will have larger coverage but less accuracy.