[
https://issues.apache.org/jira/browse/SPARK-8497?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14648126#comment-14648126
]
Xiangrui Meng commented on SPARK-8497:
--------------------------------------
Please provide the algorithm you want to implement, which should be based on
some published work for correctness. I don't know how to handle the exponential
growth of number of cliques. For example, if we have a clique of size 40, there
will be (40 choose 20) cliques of size 20, which is more than 100 billion.
> Graph Clique(Complete Connected Sub-graph) Discovery Algorithm
> --------------------------------------------------------------
>
> Key: SPARK-8497
> URL: https://issues.apache.org/jira/browse/SPARK-8497
> Project: Spark
> Issue Type: New Feature
> Components: GraphX, ML, MLlib, Spark Core
> Reporter: Fan Jiang
> Assignee: Fan Jiang
> Labels: features
> Original Estimate: 72h
> Remaining Estimate: 72h
>
> In recent years, social network industry has high demand on Complete
> Connected Sub-Graph Discoveries, so does Telecom. Similar as the graph
> connection from Twitter, the calls and other activities from telecoms world
> form a huge social graph, and due to the nature of communication method, it
> shows the strongest inter-person relationship, the graph based analysis will
> reveal tremendous value from telecoms connections.
> We need an algorithm in Spark to figure out ALL the strongest completely
> connected sub-graph (so called Clique here) for EVERY person in the network
> which will be one of the start point for understanding user's social
> behaviour.
> In Huawei, we have many real-world use cases that invovle telecom social
> graph of tens billion edges and hundreds million vertices, and the cliques
> will be also in tens million level. The graph will be a fast changing one
> which means we need to analyse the graph pattern very often (one result per
> day/week for moving time window which spans multiple months).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
