org.lenskit.rerank

Class AbstractScoringGreedyRerankStrategy

java.lang.Object

org.lenskit.rerank.AbstractScoringGreedyRerankStrategy

All Implemented Interfaces:

GreedyRerankStrategy

public abstract class AbstractScoringGreedyRerankStrategy
extends java.lang.Object
implements GreedyRerankStrategy

Abstract class designed to make implementation of a GreedyRerankStrategy easier. Implements a method of selecting the maximum scoring item that satisfies a constraint Has abstract methods for the method of choosing how many items to consider before returning the best candidate. note - While this class does have code for hard constraints, it is not a good option for implementing only a hard constraint, in those cases an AbstractFilteringGreedyRerankStrategy should be used.

Constructor Summary

Constructors

Constructor and Description
AbstractScoringGreedyRerankStrategy()

Method Summary

Modifier and Type	Method and Description
protected int	computeNumToInspect(int numRequested, int numSelected, int numCandidates) A method to compute how many of the items in the list to inspect before returning the best item found so far.
Result	nextItem(long userId, int n, java.util.List<? extends Result> items, java.util.List<? extends Result> candidates) A method to select the next item to be recommended.
protected boolean	satisfiesConstraint(long userId, int n, java.util.List<? extends Result> items, Result candidate) Tests if adding the candidate item to the list of recommended items would satisfy a constraint.
protected abstract double	scoreCandidate(long userId, int n, java.util.List<? extends Result> items, Result candidate) Computes an objective metric score for adding a given candidate item to a list of recommended items.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

AbstractScoringGreedyRerankStrategy

public AbstractScoringGreedyRerankStrategy()

Method Detail

nextItem

@Nullable
public Result nextItem(long userId,
                                 int n,
                                 java.util.List<? extends Result> items,
                                 java.util.List<? extends Result> candidates)

Description copied from interface: GreedyRerankStrategy

A method to select the next item to be recommended. This method will be called many times in the process of generating recommendations. Therefore this method should consider optimzation options such as only searching a constant number of the candidates list before picking an item.

Specified by:

nextItem in interface GreedyRerankStrategy

Parameters:

userId - the id of the user receiving these recommendation.

n - the total number of items that will be recommended.

items - The list of items already chosen to be recommended in recommendation order

candidates - A ranked list of all items in the system that can still be recommended in this context. Given in ranking order (which is not strictly guaranteed to be in score order, but often will be).

Returns:

A Result object noting which candidate item should be added to the recommendation list. The object returned will be directly added to the result list, therefore implementations should use a custom Result subclass if there is any interesting information about the recommendation process to be returned. Alternatively, this method can return a null to indicate that there is no item that can be added to the list without violating a constraint and that the recommendation list should be prematurely terminated.

scoreCandidate

protected abstract double scoreCandidate(long userId,
                                         int n,
                                         java.util.List<? extends Result> items,
                                         Result candidate)

Computes an objective metric score for adding a given candidate item to a list of recommended items. This method will only be called on items that satisfy the constraint.

Parameters:

userId - the id of the user to recommend for

n - the number of recommended items requested

items - the list of items already chosen for recommendation

candidate - the candidate item to recommend

Returns:

a score used to chose which item to recommend, with larger values considered better.

satisfiesConstraint

protected boolean satisfiesConstraint(long userId,
                                      int n,
                                      java.util.List<? extends Result> items,
                                      Result candidate)

Tests if adding the candidate item to the list of recommended items would satisfy a constraint.

The default implementation returns true, effectively disabling the constraint by default.

Parameters:

userId - the id of the user to recommend for

n - the number of recommended items requested

items - the list of items already chosen for recommendation

candidate - the candidate item to recommend

Returns:

true if adding the candidate to the recommendation would not violate the constraint.

computeNumToInspect

protected int computeNumToInspect(int numRequested,
                                  int numSelected,
                                  int numCandidates)

A method to compute how many of the items in the list to inspect before returning the best item found so far. The default implementation searches the whole list. Override this method if you want to only search some prefix of the ranking order for an optimal item.

Parameters:

numRequested - the number of items requested for recommendation

numSelected - the number of items which have already been selected

numCandidates - the number of items which can still be selected from

Returns:

the number of items to check before returning the best seen item.