-PageRank for Semi-Supervised Learning

This repository complements our manuscript entitled '-PageRank for Semi-Supervised Learning' by Esteban Bautista, Patrice Abry and Paulo Gonçalves (Submitted to Applied Network Science).

The repository provides both the results reported in the paper and the code to obtain them.

Tools

Codes are written in MATLAB.

The studied datasets are (as cited on the article): MNIST [26], Gender images [27], BBC articles [28], Phoneme [29]. They are all included in the Datasets folder.

Experiments

Validation of Algorithm 1 in estimation of the optimal

The Algorithm_evaluation folder contains:

• the evaluation of Algorithm 1 on the MNIST (reported in Table 1)

• the code to generate Figure 2

Experiments on the Planted partition

The PlantedPartition_experiment folder contains:

• the assessment of -PageRank on the Planted Partition when partitions are retrieved by means of the sweep-cut (reported in Figure 3)

• the code to generate Figure 3

Experiments on real world datasets

The RealWorldData_experiment folder contains the assessment of Algorithm 1 and -PageRank (with partitions via the sweep-cut) on the classification of real world datasets (both results reported in Table 2).

Experiments on unbalanced labeled data

The UnbalancedLabels_experiment folder contains the performance assessment (multi-class approach) of -PageRank in the presence of unbalanced labeled data (reported in Table 3).

Questions

Please contact esteban.bautista-ruiz@ens-lyon.fr for any questions or problems.