|
Title:
|
CONPREDICT: A METHOD FOR LINK PREDICTION IN CO-AUTHORED CONTENT-BASED NETWORKS |
|
Author(s):
|
Jamilson Batista Antunes, Josué Batista Antunes, Hélio Fernando Bentzen Pessoa Filho, Renato Dourado Maia, Rodrigo Barbosa de Queiroz, Carlo Marcelo Revoredo da Silva, Ricardo Batista Rodrigues, Flavi |
|
ISBN:
|
978-989-8533-16-6 |
|
Editors:
|
Bebo White and Pedro Isaías |
|
Year:
|
2013 |
|
Edition:
|
Single |
|
Keywords:
|
Link prediction, Co-author Network, Links, Similarity, Stemming |
|
Type:
|
Full Paper |
|
First Page:
|
11 |
|
Last Page:
|
18 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
Studies on link prediction aims to understand mechanisms that trigger the changes and evolution in social networks, i.e., how they grow and change over time. The evolution of a social network to let larger and more complex. This complicates the prediction of relationships. Consider a network of co-authored among scientists, for example. There are many reasons exogenous to the network, which take two scientists who have published together not to post more in the coming years, or who are very close in the network (considering the graph structure), but never published together. Such collaborations can be difficult to predict. This paper proposes a method (named ConPredict) to predict relationships in co-authored networks based on the content similarity between nodes and a prototype implementation to validate the proposed method. The ConPredict system is based on a hybrid approach (based on the structural patterns on network and similarity between nodes) to relationships predict. Experiments on large co-authored networks suggest that information about future interactions can be extracted from the topology, and the content similarity of each node in the network, and that the hybrid approach can overcome the prediction analysis only of the structural network. We applied the following measures of performance evaluation: precision, recall, F-measure. The experimental evaluation carried out gave encouraging results: precision score of 88% and 84% of recall. |
|
|
|
|
|
|
