Knowledge Discovery from Citation Networks

Code:

Download

The above code use the Lightspeed and Fastfit packages by Tom Minka.

Datasets:

Cora The original dataset is from Professor Andrew McCallum

CiteSeer The original dataset is from Professor C. Lee Giles

Detailed Experimental Results:

Topics Detection:

300 Topics obtained by BPT model on Cora corpus

Topics obtained by Link-LDA model on Cora corpus (10 topics, 50 topics, 62 topics, 300 topics)

Topics obtained by LDA model on Cora corpus (100 topics, 300 topics)

Document Recommendation according to the posterior probability within each of 300 topics for the following different model

BPT model

Link-LDA model

LDA model

Citation Recommendation by BPT model