Node2Vec

This feature is in the beta tier. For more information on feature tiers, see API Tiers.

Glossary

Directed

Directed trait. The algorithm is well-defined on a directed graph.

Directed

Directed trait. The algorithm ignores the direction of the graph.

Directed

Directed trait. The algorithm does not run on a directed graph.

Undirected

Undirected trait. The algorithm is well-defined on an undirected graph.

Undirected

Undirected trait. The algorithm ignores the undirectedness of the graph.

Heterogeneous nodes

Heterogeneous nodes fully supported. The algorithm has the ability to distinguish between nodes of different types.

Heterogeneous nodes

Heterogeneous nodes allowed. The algorithm treats all selected nodes similarly regardless of their label.

Heterogeneous relationships

Heterogeneous relationships fully supported. The algorithm has the ability to distinguish between relationships of different types.

Heterogeneous relationships

Heterogeneous relationships allowed. The algorithm treats all selected relationships similarly regardless of their type.

Weighted relationships

Weighted trait. The algorithm supports a relationship property to be used as weight, specified via the relationshipWeightProperty configuration parameter.

Weighted relationships

Weighted trait. The algorithm treats each relationship as equally important, discarding the value of any relationship weight.

Node2Vec is a node embedding algorithm that computes a vector representation of a node based on random walks in the graph. The neighborhood is sampled through random walks. Using a number of random neighborhood samples, the algorithm trains a single hidden layer neural network. The neural network is trained to predict the likelihood that a node will occur in a walk based on the occurrence of another node.

For more information on this algorithm, see:

Random Walks

A main concept of the Node2Vec algorithm are the second order random walks. A random walk simulates a traversal of the graph in which the traversed relationships are chosen at random. In a classic random walk, each relationship has the same, possibly weighted, probability of being picked. This probability is not influenced by the previously visited nodes. The concept of second order random walks, however, tries to model the transition probability based on the currently visited node v, the node t visited before the current one, and the node x which is the target of a candidate relationship. Node2Vec random walks are thus influenced by two parameters: the returnFactor and the inOutFactor:

  • The returnFactor is used if t equals x, i.e., the random walk returns to the previously visited node.

  • The inOutFactor is used if the distance from t to x is equal to 2, i.e., the walk traverses further away from the node t

Visuzalition of random walk parameters

The probabilities for traversing a relationship during a random walk can be further influenced by specifying a relationshipWeightProperty. A relationship property value greater than 1 will increase the likelihood of a relationship being traversed, a property value between 0 and 1 will decrease that probability.

For every node in the graph Node2Vec generates a series of random walks with the particular node as start node. The number of random walks per node can be influenced by the walkPerNode configuration parameters, the walk length is controlled by the walkLength parameter.

Usage in machine learning pipelines

At this time, using Node2Vec as a node property step in a machine learning pipeline (like Link prediction pipelines and Node property prediction) is not well supported, at least if the end goal is to apply a prediction model using its embeddings.

In order for a machine learning model to be able to make useful predictions, it is important that features produced during prediction are of a similar distribution to the features produced during training of the model. Moreover, node property steps (whether Node2Vec or not) added to a pipeline are executed both during training, and during the prediction by the trained model. It is therefore problematic when a pipeline contains an embedding step which yields all too dissimilar embeddings during training and prediction.

The final embeddings produced by Node2Vec depends on the randomness in generating the initial node embedding vectors as well as the random walks taken in the computation. At this time, Node2Vec will produce non-deterministic results even if the randomSeed configuration parameter is set. So since embeddings will not be deterministic between runs, Node2Vec should not be used as a node property step in a pipeline at this time, unless the purpose is experimental and only the train mode is used.

It may still be useful to use Node2Vec node embeddings as features in a pipeline if they are produced outside the pipeline, as long as one is aware of the data leakage risks of not using the dataset split in the pipeline.

Syntax

Node2Vec syntax per mode
Run Node2Vec in stream mode on a named graph.
CALL gds.node2vec.stream(
  graphName: String,
  configuration: Map
) YIELD
  nodeId: Integer,
  embedding: List of Float
Table 1. Parameters
Name Type Default Optional Description

graphName

String

n/a

no

The name of a graph stored in the catalog.

configuration

Map

{}

yes

Configuration for algorithm-specifics and/or graph filtering.

Table 2. Configuration
Name Type Default Optional Description

nodeLabels

List of String

['*']

yes

Filter the named graph using the given node labels. Nodes with any of the given labels will be included.

relationshipTypes

List of String

['*']

yes

Filter the named graph using the given relationship types. Relationships with any of the given types will be included.

concurrency

Integer

4

yes

The number of concurrent threads used for running the algorithm.

jobId

String

Generated internally

yes

An ID that can be provided to more easily track the algorithm’s progress.

logProgress

Boolean

true

yes

If disabled the progress percentage will not be logged.

walkLength

Integer

80

yes

The number of steps in a single random walk.

walksPerNode

Integer

10

yes

The number of random walks generated for each node.

inOutFactor

Float

1.0

yes

Tendency of the random walk to stay close to the start node or fan out in the graph. Higher value means stay local.

returnFactor

Float

1.0

yes

Tendency of the random walk to return to the last visited node. A value below 1.0 means a higher tendency.

relationshipWeightProperty

String

null

yes

Name of the relationship property to use as weights to influence the probabilities of the random walks. The weights need to be >= 0. If unspecified, the algorithm runs unweighted.

windowSize

Integer

10

yes

Size of the context window when training the neural network.

negativeSamplingRate

Integer

5

yes

Number of negative samples to produce for each positive sample.

positiveSamplingFactor

Float

0.001

yes

Factor for influencing the distribution for positive samples. A higher value increases the probability that frequent nodes are down-sampled.

negativeSamplingExponent

Float

0.75

yes

Exponent applied to the node frequency to obtain the negative sampling distribution. A value of 1.0 samples proportionally to the frequency. A value of 0.0 samples each node equally.

embeddingDimension

Integer

128

yes

Size of the computed node embeddings.

embeddingInitializer

String

NORMALIZED

yes

Method to initialize embeddings. Values are sampled uniformly from a range [-a, a]. With NORMALIZED, a=0.5/embeddingDimension and with UNIFORM instead a=1.

iterations

Integer

1

yes

Number of training iterations.

initialLearningRate

Float

0.01

yes

Learning rate used initially for training the neural network. The learning rate decreases after each training iteration.

minLearningRate

Float

0.0001

yes

Lower bound for learning rate as it is decreased during training.

randomSeed

Integer

random

yes

Seed value used to generate the random walks, which are used as the training set of the neural network. Note, that the generated embeddings are still nondeterministic.

walkBufferSize

Integer

1000

yes

The number of random walks to complete before starting training.

Table 3. Results
Name Type Description

nodeId

Integer

The Neo4j node ID.

embedding

List of Float

The computed node embedding.

Run Node2Vec in mutate mode on a graph stored in the catalog.
CALL gds.node2vec.mutate(
  graphName: String,
  configuration: Map
)
YIELD
  preProcessingMillis: Integer,
  computeMillis: Integer,
  postProcessingMillis: Integer,
  mutateMillis: Integer,
  nodeCount: Integer,
  nodePropertiesWritten: Integer,
  lossPerIteration: List of Float,
  configuration: Map
Table 4. Parameters
Name Type Default Optional Description

graphName

String

n/a

no

The name of a graph stored in the catalog.

configuration

Map

{}

yes

Configuration for algorithm-specifics and/or graph filtering.

Table 5. Configuration
Name Type Default Optional Description

mutateProperty

String

n/a

no

The node property in the GDS graph to which the embedding is written.

nodeLabels

List of String

['*']

yes

Filter the named graph using the given node labels.

relationshipTypes

List of String

['*']

yes

Filter the named graph using the given relationship types.

concurrency

Integer

4

yes

The number of concurrent threads used for running the algorithm.

jobId

String

Generated internally

yes

An ID that can be provided to more easily track the algorithm’s progress.

walkLength

Integer

80

yes

The number of steps in a single random walk.

walksPerNode

Integer

10

yes

The number of random walks generated for each node.

inOutFactor

Float

1.0

yes

Tendency of the random walk to stay close to the start node or fan out in the graph. Higher value means stay local.

returnFactor

Float

1.0

yes

Tendency of the random walk to return to the last visited node. A value below 1.0 means a higher tendency.

relationshipWeightProperty

String

null

yes

Name of the relationship property to use as weights to influence the probabilities of the random walks. The weights need to be >= 0. If unspecified, the algorithm runs unweighted.

windowSize

Integer

10

yes

Size of the context window when training the neural network.

negativeSamplingRate

Integer

5

yes

Number of negative samples to produce for each positive sample.

positiveSamplingFactor

Float

0.001

yes

Factor for influencing the distribution for positive samples. A higher value increases the probability that frequent nodes are down-sampled.

negativeSamplingExponent

Float

0.75

yes

Exponent applied to the node frequency to obtain the negative sampling distribution. A value of 1.0 samples proportionally to the frequency. A value of 0.0 samples each node equally.

embeddingDimension

Integer

128

yes

Size of the computed node embeddings.

embeddingInitializer

String

NORMALIZED

yes

Method to initialize embeddings. Values are sampled uniformly from a range [-a, a]. With NORMALIZED, a=0.5/embeddingDimension and with UNIFORM instead a=1.

iterations

Integer

1

yes

Number of training iterations.

initialLearningRate

Float

0.01

yes

Learning rate used initially for training the neural network. The learning rate decreases after each training iteration.

minLearningRate

Float

0.0001

yes

Lower bound for learning rate as it is decreased during training.

randomSeed

Integer

random

yes

Seed value used to generate the random walks, which are used as the training set of the neural network. Note, that the generated embeddings are still nondeterministic.

walkBufferSize

Integer

1000

yes

The number of random walks to complete before starting training.

Table 6. Results
Name Type Description

nodeCount

Integer

The number of nodes processed.

nodePropertiesWritten

Integer

The number of node properties written.

preProcessingMillis

Integer

Milliseconds for preprocessing the data.

computeMillis

Integer

Milliseconds for running the algorithm.

mutateMillis

Integer

Milliseconds for adding properties to the projected graph.

postProcessingMillis

Integer

Milliseconds for post-processing of the results.

lossPerIteration

List of Float

The sum of the losses registered per training iteration.

configuration

Map

The configuration used for running the algorithm.

Run Node2Vec in write mode on a graph stored in the catalog.
CALL gds.node2vec.write(
  graphName: String,
  configuration: Map
)
YIELD
  preProcessingMillis: Integer,
  computeMillis: Integer,
  writeMillis: Integer,
  nodeCount: Integer,
  nodePropertiesWritten: Integer,
  lossPerIteration: List of Float,
  configuration: Map
Table 7. Parameters
Name Type Default Optional Description

graphName

String

n/a

no

The name of a graph stored in the catalog.

configuration

Map

{}

yes

Configuration for algorithm-specifics and/or graph filtering.

Table 8. Configuration
Name Type Default Optional Description

nodeLabels

List of String

['*']

yes

Filter the named graph using the given node labels. Nodes with any of the given labels will be included.

relationshipTypes

List of String

['*']

yes

Filter the named graph using the given relationship types. Relationships with any of the given types will be included.

concurrency

Integer

4

yes

The number of concurrent threads used for running the algorithm.

jobId

String

Generated internally

yes

An ID that can be provided to more easily track the algorithm’s progress.

logProgress

Boolean

true

yes

If disabled the progress percentage will not be logged.

writeConcurrency

Integer

value of 'concurrency'

yes

The number of concurrent threads used for writing the result to Neo4j.

writeProperty

String

n/a

no

The node property in the Neo4j database to which the embedding is written.

walkLength

Integer

80

yes

The number of steps in a single random walk.

walksPerNode

Integer

10

yes

The number of random walks generated for each node.

inOutFactor

Float

1.0

yes

Tendency of the random walk to stay close to the start node or fan out in the graph. Higher value means stay local.

returnFactor

Float

1.0

yes

Tendency of the random walk to return to the last visited node. A value below 1.0 means a higher tendency.

relationshipWeightProperty

String

null

yes

Name of the relationship property to use as weights to influence the probabilities of the random walks. The weights need to be >= 0. If unspecified, the algorithm runs unweighted.

windowSize

Integer

10

yes

Size of the context window when training the neural network.

negativeSamplingRate

Integer

5

yes

Number of negative samples to produce for each positive sample.

positiveSamplingFactor

Float

0.001

yes

Factor for influencing the distribution for positive samples. A higher value increases the probability that frequent nodes are down-sampled.

negativeSamplingExponent

Float

0.75

yes

Exponent applied to the node frequency to obtain the negative sampling distribution. A value of 1.0 samples proportionally to the frequency. A value of 0.0 samples each node equally.

embeddingDimension

Integer

128

yes

Size of the computed node embeddings.

embeddingInitializer

String

NORMALIZED

yes

Method to initialize embeddings. Values are sampled uniformly from a range [-a, a]. With NORMALIZED, a=0.5/embeddingDimension and with UNIFORM instead a=1.

iterations

Integer

1

yes

Number of training iterations.

initialLearningRate

Float

0.01

yes

Learning rate used initially for training the neural network. The learning rate decreases after each training iteration.

minLearningRate

Float

0.0001

yes

Lower bound for learning rate as it is decreased during training.

randomSeed

Integer

random

yes

Seed value used to generate the random walks, which are used as the training set of the neural network. Note, that the generated embeddings are still nondeterministic.

walkBufferSize

Integer

1000

yes

The number of random walks to complete before starting training.

Table 9. Results
Name Type Description

nodeCount

Integer

The number of nodes processed.

nodePropertiesWritten

Integer

The number of node properties written.

preProcessingMillis

Integer

Milliseconds for preprocessing the data.

computeMillis

Integer

Milliseconds for running the algorithm.

writeMillis

Integer

Milliseconds for writing result data back to Neo4j.

lossPerIteration

List of Float

The sum of the losses registered per training iteration.

configuration

Map

The configuration used for running the algorithm.

Examples

All the examples below should be run in an empty database.

The examples use native projections as the norm, although Cypher projections can be used as well.

Consider the graph created by the following Cypher statement:

CREATE (alice:Person {name: 'Alice'})
CREATE (bob:Person {name: 'Bob'})
CREATE (carol:Person {name: 'Carol'})
CREATE (dave:Person {name: 'Dave'})
CREATE (eve:Person {name: 'Eve'})
CREATE (guitar:Instrument {name: 'Guitar'})
CREATE (synth:Instrument {name: 'Synthesizer'})
CREATE (bongos:Instrument {name: 'Bongos'})
CREATE (trumpet:Instrument {name: 'Trumpet'})

CREATE (alice)-[:LIKES]->(guitar)
CREATE (alice)-[:LIKES]->(synth)
CREATE (alice)-[:LIKES]->(bongos)
CREATE (bob)-[:LIKES]->(guitar)
CREATE (bob)-[:LIKES]->(synth)
CREATE (carol)-[:LIKES]->(bongos)
CREATE (dave)-[:LIKES]->(guitar)
CREATE (dave)-[:LIKES]->(synth)
CREATE (dave)-[:LIKES]->(bongos);
CALL gds.graph.project('myGraph', ['Person', 'Instrument'], 'LIKES');
Run the Node2Vec algorithm on myGraph
CALL gds.node2vec.stream('myGraph', {embeddingDimension: 2})
YIELD nodeId, embedding
RETURN nodeId, embedding
Table 10. Results
nodeId embedding

0

[-0.14295829832553864, 0.08884537220001221]

1

[0.016700705513358116, 0.2253911793231964]

2

[-0.06589698046445847, 0.042405471205711365]

3

[0.05862073227763176, 0.1193704605102539]

4

[0.10888434946537018, -0.18204474449157715]

5

[0.16728264093399048, 0.14098615944385529]

6

[-0.007779224775731564, 0.02114257402718067]

7

[-0.213893860578537, 0.06195802614092827]

8

[0.2479933649301529, -0.137322798371315]