LPCNN: convolutional neural network for link prediction based on network structured features

• Published in: Telkomnika 2022-12, Vol.20 (6), p.1214-1224
• Main Authors: Alzubaidi, Asia Mahdi Naser, Alsaadi, Elham Mohammed Thabit A.
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Artificial neural networks; Decision trees; Deep learning; Graph theory; Heuristic; Heuristic methods; Machine learning; Network analysis; Network topologies; Neural networks; Proteins; Social network analysis; Social networks; Support vector machines
• ISSN: 1693-6930; 2302-9293
• DOI: 10.12928/telkomnika.v20i6.22990

First-order heuristics like common neighbors and preferred attachment only contain one-hop neighbors of two chosen nodes. [...]high-order heuristic methods frequently outperform low-order heuristic approaches, although they have a higher computational cost. Because numerous heuristic techniques have been developed to handle various graphs, finding a suitable heuristic approach becomes a difficult task [10]. [...]embedding algorithms that can learn node features from network topology have been employed to resolve the LP issue; notable approaches in this line include matrix factorization and stochastic block modeling (SBM) [12]. Predicting links by analyzing common neighbors (PLACN) a methodology based on convolutional neural networks is introduced and compared their technique to the state-of-the-art method, achieving 96% area under curve (AUC) in the benchmark [18]. Because of its accuracy, a subgraph technique known as Weisfeiler-Lehman neural machine (WLNM) was recently designated as a state-of-the-art link prediction method [19].