PersonalRank

Overview

PersonalRank is also called Personalized PageRank. It inherits the idea of the classic PageRank algorithm and uses the graph link structure to recursively calculate the importance of each node. However, unlike the PageRank algorithm, to ensure that the access probability of each node in the random walk can reflect user preferences, the PersonalRank algorithm returns each hop to the source node at a (1-alpha) probability during random walk. Therefore, the relevance and importance of network nodes can be calculated based on the source node. (The higher the PersonalRank value, the higher the correlation/importance of the source node.)

Application Scenarios

This algorithm applies to fields such as commodity, friend, and web page recommendations.

Parameter Description

**Table 1** PersonalRank algorithm parameters
Parameter	Mandatory	Description	Type	Value Range	Default Value
source	Yes	Node ID	String	-	-
alpha	No	Weight coefficient	Double	A real number between 0 and 1 (excluding 0 and 1)	0.85
convergence	No	Convergence	Double	A real number between 0 and 1 (excluding 0 and 1)	0.00001
max_iterations	No	Maximum iterations	Int	1-2,000	1,000
directed	No	Whether to consider the edge direction	Bool	true or false	true

alpha determines the jump probability coefficient, also called damping coefficient, which is a computing control variable in the algorithm.
convergence defines the sum and upper limit of absolute values of each vertex in each iteration compared with the last iteration. If the sum is less than the value, the computing is considered to be converged and the algorithm stops.

Precautions

When the convergence is set to a large value, the iteration will stop quickly.

Example

Set parameters source to Lee, alpha to 0.85, convergence to 0.00001, max_iterations to 1,000, and directed to true. The sub-graph formed by top nodes in the calculation result is displayed on the canvas. The size of a node varies with the PersonalRank values. The JSON result is displayed in the query result area.