UPDATE: you can find an evaluation of the following blog post and idea on: http://www.rene-pickhardt.de/graphity-an-efficient-graph-model-for-retrieving-the-top-k-news-feeds-for-users-in-social-networks/
Hey everyone I can finally demonstrate the neo4j and gwt system that I have been blogging about over the last weeks here and here. But please find the demo under the following adress:
http://gwt.metalcon.de/GWT-Modelling
The code will be available soon! But it was hacked together pretty nasty and definatly needs some cleanup! Also I think that I still have one lack in memory while maintinging the index for my news feed. Hopefully I will be able to do this while writing my paper and do more on the evaluation. Already thanks a lot to Jonas Kunze and Dr. Thomas Gottron as well as Peter from neo technologies for the great support during this project!
The Graph that I used is extracted from the complete revision dump of the bavarian wikipedia and has the following properties:
- 8’760 entities
- ~95’000 Content items
- ==> together more than 100’000 nodes
- almost 20’000 different relationship types (more to come in bigger graphs!)
- about 100’000 edges connecting the 8’760 entities
- mainting the index was possible with (only :[ ) 177 writes / second on a slow notebook computer
I interpret the wikipedia graph in the following way as a social network:
- every article corresponds to an entity
- every link corresponds to a directed friendship (follower)
- every revision corresponds eather to a status update (content item) or a change in the friendship graph
I have no idea yet how many reads I can do per second. Even though I am a little disappointed about the low speed for writing on the graph I am sure that I will achieve my theoretical goals for reads per second. I also hope to increase writing spead if I introduce better transaction management. Anyway I will blog about the results of the reads / second later.