1: \begin{abstract}

2: 

3: An enhanced label propagation (LP) method called GraphHop was proposed

4: recently. It outperforms graph convolutional networks (GCNs) in the

5: semi-supervised node classification task on various networks.  Although

6: the performance of GraphHop was explained intuitively with joint node

7: attribute and label signal smoothening, its rigorous mathematical treatment

8: is lacking. In this paper, we propose a label efficient regularization

9: and propagation (LERP) framework for graph node classification, and

10: present an alternate optimization procedure for its solution.

11: Furthermore, we show that GraphHop only offers an approximate solution

12: to this framework and has two drawbacks. First, it includes all nodes in

13: the classifier training without taking the reliability of pseudo-labeled

14: nodes into account in the label update step.  Second, it provides a

15: rough approximation to the optimum of a subproblem in the label

16: aggregation step. Based on the LERP framework, we propose a new method,

17: named the LERP method, to solve these two shortcomings. LERP determines

18: reliable pseudo-labels adaptively during the alternate optimization and

19: provides a better approximation to the optimum with computational

20: efficiency.  Theoretical convergence of LERP is guaranteed.  Extensive

21: experiments are conducted to demonstrate the effectiveness and

22: efficiency of LERP. That is, LERP outperforms all benchmarking methods,

23: including GraphHop, consistently on five test datasets and an object

24: recognition task at extremely low label rates (i.e., 1, 2, 4, 8, 16, and

25: 20 labeled samples per class). 

26: 

27: \end{abstract}

28: