Supervised-learning link recommendation in the DBLP co-authoring network

Abstract

Currently, link recommendation has gained more attention as networked data becomes abundant in several scenarios. However, existing methods for this task have failed in considering solely the structure of dynamic networks for improved performance and accuracy. Hence, in this work, we present a methodology based on the use of multiple topological metrics in order to achieve prospective link recommendations considering time constraints. The combination of such metrics is used as input to binary classification algorithms that state whether two pairs of authors will/should define a link. We experimented with five algorithms, what allowed us to reach high rates of accuracy and to evaluate the different classification paradigms. Our results also demonstrated that time parameters and the activity profile of the authors can significantly influence the recommendation. In the context of DBLP, this research is strategic as it may assist on identifying potential partners, research groups with similar themes, research competition (absence of obvious links), and related work.

HeteroRWR: A Novel Algorithm for Top-<i>k</i> Co-Author Recommendation with Fusion of Citation Networks

“…From the prospective of learning to rank [23], supervised friend recommendation approaches can be divided into pointwise learning, pairwise learning and listwise learning methods. Pointwise learning models transfer the sorting problem into a multi-classification problem or a regression problem [11], [12], and the disadvantage is that these models cannot deal with the high skewness of data very well. Pairwise learning models treat friend recommendation as a learning to rank problem based upon pairwise comparisons [14], [19], [24]- [26], [43].…”

“…Next, we select 3 − core authors [11] from all authors, i.e. the authors who had published at least three papers during the training and the test intervals, respectively.…”

Multimodal graph-based analysis over the DBLP repository

“…We do this by means of link recommendation techniques together with measures of accuracy. Here we verify this property by extracting metrics from the co-authoring network of DBLP; specifically, for each author we extract Number of common authors (a), Jaccard's coefficient (b), Preferential attachment (c), Adamic-Adar coefficient (d), Resource allocation index (f), and Local path (g) -please check the work of Gimenes et al [7] for details. These metrics are then used with supervised machine learning classifiers [19] J48, Naïve Bayes, Multilayer Perceptron, Bagging, and Random Forest, all of them available in the Weka framework, developed by the University of Waikato [5].…”

Mining user behavior in location-based social networks

“…Some authors have compared different traditional classifiers including decision trees, support vector machines, k-nearest neighbors, multi-layer perceptrons, naïve Bayes, and others, as well as different ensembles of these classifiers (HASAN et al, 2006;BENCHETTARA;KANAWATI;ROUVEIROL, 2010;LUSSIER;CHAWLA, 2010;FIRE et al, 2011;FIRE et al, 2014;GIMENES et al, 2014). Other authors have obtained good results using random forest classifiers (CUKIERSKI; HAMNER; YANG, 2011), or using classifiers The main challenge that link prediction methods based on classifiers has to deal with is the well-known class imbalance problem.…”