hidden abstract class %iFind.Rank.Abstract extends %Library.RegisteredObject
This is the abstract superclass for all ranking algorithm implementations. In order to
provide a rank or score for one or more records, an instance of this class is created and
initialized with the search string and options. As part of this initialization, the search
string will be parsed in the same way as it's done for the search operation itself and the
resulting search tree is stored in the ParsedQuery.
Implementation-specific setup can be carried out by overriding the OnInitialize() method, for example to prepare term weights and other reusable metrics that are common for all per-record calculations.
Then, the GetRecordScore() method is invoked for every record, first invoking its OnBeforeGetRecordScore() and then traversing the search tree's nodes one by one through calls to GetNodeScore(), skipping the ones that are exclusion criteria (NOT). For each leaf node, GetLeafScore() is invoked to obtain a score expressing how well that record matches that leaf node's search string.
These scores are then aggregated by taking the minimum value when nodes are ANDed and the maximum value when nodes are ORed, producing an overall score for the record, which can finally be amended through the OnGetRecordScore() callback.
Subclasses should typically only override / implement the GetLeafScore() and do any preparation or cleanup in the callback methods.
Defines how weight scores of different subnodes of an AND node in the search tree are combined by the GetNodeScore() method to provide an aggregated weight for the total. See OPERATORLOGICOR for a list of options.
Defines how weight scores of different subnodes of an OR node in the search tree are combined by the GetNodeScore() method to provide an aggregated weight for the total. The different values are:
- MAX: the highest weight of any sub-node is used as the weight of this composite node
- MIN: the smallest weight of any sub-node is used as the weight of this composite node
- SUM: the sum of all sub-node's weights is used as the weight of this composite node
- AVG: the average of all sub-node's weights is used as the weight of this composite node
ParsedQuery([node ID], "o") = [operator: A|O|N] ParsedQuery([node ID], "p") = [parent node ID] ParsedQuery([node ID], "ch", [child node ID]) = [child node is negated] ParsedQuery([node ID], "s") = n ParsedQuery([node ID], "s", [leaf ID]) = [atomic string] ParsedQuery([node ID], "s", [leaf ID], "n") = [atomic string is negated] ParsedQuery([node ID], "s", [leaf ID], 1) = [left operator] ParsedQuery([node ID], "s", [leaf ID], 2) = [right operator]
This method calculates the overall score for the given search string for one specific record by invoking GetNodeScore() on the top-level node of the parsed search string. The resulting score can still be amended by OnGetRecordScore(), order dto apply any record-level modifiers.
This method will also invoke the callback methods OnBeforeGetRecordScore() and OnAfterGetRecordScore(), which can be used to retrieve and clean up any record- related information required for the GetLeafScore() implementation.
OnBeforeGetRecordScore() implemenations may return a skip flag, instructing this code to skip the calls to GetNodeScore() for this record.