Package eu.h2020.helios_social.modules.socialgraphmining.diffusion

Class PPRMiner

java.lang.Object

eu.h2020.helios_social.modules.socialgraphmining.SocialGraphMiner

eu.h2020.helios_social.modules.socialgraphmining.diffusion.PPRMiner

public class PPRMiner
extends SocialGraphMiner

This class implements a Personalized PageRank scheme, where each ego node's personalization is a vector (modeled by a JGNN library Tensor) passed to the constructor and which is smoothed over the decentralized social graph. Smoothing outcome for contexts is obtained through the method getSmoothedPersonalization(Context). The miner predicts new interactions based on the cosine similarity of smoothed attributes.

Author:

Emmanouil Krasanakis

Nested Class Summary

Nested classes/interfaces inherited from class eu.h2020.helios_social.modules.socialgraphmining.SocialGraphMiner

SocialGraphMiner.InteractionType, SocialGraphMiner.SocialGraphMinerParameters

Constructor Summary

Constructors

Constructor	Description
PPRMiner(java.lang.String name, eu.h2020.helios_social.core.contextualegonetwork.ContextualEgoNetwork contextualEgoNetwork, mklab.JGNN.core.Tensor personalization)

Method Summary

Modifier and Type	Method	Description
SocialGraphMiner.SocialGraphMinerParameters	constructModelParameterObject(eu.h2020.helios_social.core.contextualegonetwork.Interaction interaction)
java.lang.String	getName()
mklab.JGNN.core.Tensor	getNormalizedSmoothedPersonalization(eu.h2020.helios_social.core.contextualegonetwork.Context context)	Retrieves the outcome of getSmoothedPersonalization(Context) and postprocesses it so that its minimum value is zero and its elements sum to 1.
mklab.JGNN.core.Tensor	getPersonalization(eu.h2020.helios_social.core.contextualegonetwork.Context context)	Retrieves the ego node's personalization set for the specific context.
double	getRestartProbability()	Retrieves the restart probability of the equivalent random walk with restart scheme associated with personalized PageRank.
mklab.JGNN.core.Tensor	getSmoothedPersonalization(eu.h2020.helios_social.core.contextualegonetwork.Context context)	Retrieves the outcome of smoothing the outcome of the ego's personalization through the social graph.
void	newInteractionParameters(eu.h2020.helios_social.core.contextualegonetwork.Interaction interaction, SocialGraphMiner.SocialGraphMinerParameters neighborModelParameters, SocialGraphMiner.InteractionType interactionType)
double	predictNewInteraction(eu.h2020.helios_social.core.contextualegonetwork.Context context, eu.h2020.helios_social.core.contextualegonetwork.Node destinationNode)	Predicts the weight of performing a SEND interaction between the given context's ego and a destination node within a given context.
PPRMiner	setPersonalizationAsGroundTruth(boolean personalizationAsGroundTruth)	Sets whether personalization should be considered as ground truth.
PPRMiner	setRestartProbability(double restartProbability)	Sets the restart probability of the personalized PageRank scheme.
PPRMiner	updatePersonalization(eu.h2020.helios_social.core.contextualegonetwork.Context context, mklab.JGNN.core.Tensor personalization)	Copies the elements of a given vector to the ego's personalization vector in the provided context.
PPRMiner	updatePersonalization(mklab.JGNN.core.Tensor personalization)	Wraps the method updatePersonalization(Context, Tensor) to update all contexts with the given personalization.

Methods inherited from class java.lang.Object

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

Methods inherited from class eu.h2020.helios_social.modules.socialgraphmining.SocialGraphMiner

getContextualEgoNetwork, getModelParameterObject, getModelParameters, newInteraction, recommendInteractions, setSendPermision

Constructor Detail

PPRMiner

public PPRMiner(java.lang.String name,
                eu.h2020.helios_social.core.contextualegonetwork.ContextualEgoNetwork contextualEgoNetwork,
                mklab.JGNN.core.Tensor personalization)

Method Detail

getName

public java.lang.String getName()

getRestartProbability

public double getRestartProbability()

Retrieves the restart probability of the equivalent random walk with restart scheme associated with personalized PageRank.

Returns:

The restart probability.

See Also:

setRestartProbability(double)

setPersonalizationAsGroundTruth

public PPRMiner setPersonalizationAsGroundTruth(boolean personalizationAsGroundTruth)

Sets whether personalization should be considered as ground truth. If this is true and the personalization vector has a non-zero norm (i.e. has at least one non-zero element), then the outcome of getSmoothedPersonalization(Context) is forcefully snapped to the personalization vector. Making this depend on the norm helps deployment of models.

Parameters:

personalizationAsGroundTruth - A boolean value on whether personalization should be considered ground truth.

Returns:

this miner's instance.

setRestartProbability

public PPRMiner setRestartProbability(double restartProbability)

Sets the restart probability of the personalized PageRank scheme. Smaller values induce broader diffusion of predictions, i.e. many hops away in the social graph. The equivalent random walk with restart scheme has average random walk length equal to 1/(restart probability).
Suggested values to experiment with:
- 0.15 (used in older personalized PageRank papers)
- 0.10 (default, used by graph neural networks to great success)
- 0.01 (extremely long walks, suitable to detect communities of high radius from few examples)

Parameters:

restartProbability - The restart probability in the range (0,1) (default is 0.1).

Returns:

this miner's instance.

updatePersonalization

public PPRMiner updatePersonalization(eu.h2020.helios_social.core.contextualegonetwork.Context context,
                                      mklab.JGNN.core.Tensor personalization)

Copies the elements of a given vector to the ego's personalization vector in the provided context.

Parameters:

context - The context in which to set the new personalization vector.

personalization - The tensor to set as new personalization vector.

Returns:

this miner's instance.

See Also:

updatePersonalization(Tensor)

updatePersonalization

public PPRMiner updatePersonalization(mklab.JGNN.core.Tensor personalization)

Wraps the method updatePersonalization(Context, Tensor) to update all contexts with the given personalization. The given personalization is also automatically initialized for future context generations.

Parameters:

personalization - The tensor to set as new personalization vector in all contexts.

Returns:

this miner's instance.

getPersonalization

public mklab.JGNN.core.Tensor getPersonalization(eu.h2020.helios_social.core.contextualegonetwork.Context context)

Retrieves the ego node's personalization set for the specific context. Changing the returned tensor also changes the personalization for the specific context, equivalently to updatePersonalization(Context, Tensor). However it does not affect the default personalization. Future calls to updatePersonalization(Tensor) overwrite setting custom context personalizations. This means that the PPRMiner's constructor also sets all context personalization to the same values.

Parameters:

context - The context from which to obtain the personalization.

Returns:

The personalization vector.

See Also:

getSmoothedPersonalization(Context)

getSmoothedPersonalization

public mklab.JGNN.core.Tensor getSmoothedPersonalization(eu.h2020.helios_social.core.contextualegonetwork.Context context)

Retrieves the outcome of smoothing the outcome of the ego's personalization through the social graph. Changing the output tensor temporarily affects the personalization.

Parameters:

context - The context from which to obtain the smoothed personalization.

Returns:

A Tensor holding a smoothing of the personalization.

See Also:

getPersonalization(Context), getNormalizedSmoothedPersonalization(Context)

getNormalizedSmoothedPersonalization

public mklab.JGNN.core.Tensor getNormalizedSmoothedPersonalization(eu.h2020.helios_social.core.contextualegonetwork.Context context)

Retrieves the outcome of getSmoothedPersonalization(Context) and postprocesses it so that its minimum value is zero and its elements sum to 1.

Parameters:

context - The context from which to obtain the smoothed personalization.

Returns:

A Tensor holding a normalized version of the smoothed personalization.

See Also:

getSmoothedPersonalization(Context)

newInteractionParameters

public void newInteractionParameters(eu.h2020.helios_social.core.contextualegonetwork.Interaction interaction,
                                     SocialGraphMiner.SocialGraphMinerParameters neighborModelParameters,
                                     SocialGraphMiner.InteractionType interactionType)

Specified by:

newInteractionParameters in class SocialGraphMiner

constructModelParameterObject

public SocialGraphMiner.SocialGraphMinerParameters constructModelParameterObject(eu.h2020.helios_social.core.contextualegonetwork.Interaction interaction)

predictNewInteraction

public double predictNewInteraction(eu.h2020.helios_social.core.contextualegonetwork.Context context,
                                    eu.h2020.helios_social.core.contextualegonetwork.Node destinationNode)

Description copied from class: SocialGraphMiner

Predicts the weight of performing a SEND interaction between the given context's ego and a destination node within a given context. This method should typically return values in the [0,1] range, where higher values correspond to more likely interactions.

Specified by:

predictNewInteraction in class SocialGraphMiner

Parameters:

context - The context in which to perform the prediction.

destinationNode - The destination node of the interaction.

Returns:

The weight of performing the interaction (higher values are more likely).