Complex data community detection
Posted: 02 Dec 2017 12:30
Hello,
I'm a student at the department of Informatics of Rijeka and as my final thesis I have a theme to analize the complex network of hyperlinks on the wikipedia site. I managed to get the links and form a complex network. The files are saved in .gexf format structure and when loaded they are showing a network of around 10,000 + nodes and 15,000+ edges.
As I understood, Gephi is built on the Louvain algorithm of detection. I wrote a code in Python using the Community package to detect the communities.
The problem I get is this: when I detect the number of communities through my algorithm, the number is most likely the same(small variations), but detecting through Gephi gives me some strange results(big variations).
E.g. I'm detecting communities for a directed network with 11,745 nodes and 23,255 edges.
The results I get through my algorithm are (10 testings): 115, 115, 115, 116, 115, 112, 115, 111, 114, 115.
The results I get through Gephi are (10 testings): 115, 107, 106, 108, 107, 86, 84, 112, 113, 99.
I didn't tick randomize for better decomposition and didn't tick edge weight cause I'm working on a directed unweighted graph. The resolution is set to 1(default).
If I tick any combination of those above, I still get some strange variations.
I'm not sure in my algorithm, since I wrote it as an example through a pseudo code I found on the site of the Community package.
Although the results I get seems a bit more stable(not telling they are correct). I'm missing the knowledge about the communities and detection still, so I can't be the judge if the results I get are always good/not good.
Could anyone give me an advice or opinion about this?
Btw. thank you for the constant and great work you put in Gephi! =)
I'm a student at the department of Informatics of Rijeka and as my final thesis I have a theme to analize the complex network of hyperlinks on the wikipedia site. I managed to get the links and form a complex network. The files are saved in .gexf format structure and when loaded they are showing a network of around 10,000 + nodes and 15,000+ edges.
As I understood, Gephi is built on the Louvain algorithm of detection. I wrote a code in Python using the Community package to detect the communities.
The problem I get is this: when I detect the number of communities through my algorithm, the number is most likely the same(small variations), but detecting through Gephi gives me some strange results(big variations).
E.g. I'm detecting communities for a directed network with 11,745 nodes and 23,255 edges.
The results I get through my algorithm are (10 testings): 115, 115, 115, 116, 115, 112, 115, 111, 114, 115.
The results I get through Gephi are (10 testings): 115, 107, 106, 108, 107, 86, 84, 112, 113, 99.
I didn't tick randomize for better decomposition and didn't tick edge weight cause I'm working on a directed unweighted graph. The resolution is set to 1(default).
If I tick any combination of those above, I still get some strange variations.
I'm not sure in my algorithm, since I wrote it as an example through a pseudo code I found on the site of the Community package.
Although the results I get seems a bit more stable(not telling they are correct). I'm missing the knowledge about the communities and detection still, so I can't be the judge if the results I get are always good/not good.
Could anyone give me an advice or opinion about this?
Btw. thank you for the constant and great work you put in Gephi! =)