# Line Graph Neural Networks for Link Weight Prediction

arXiv: https://arxiv.org/abs/2309.15728 · HTML: https://mlsanity.com/paper/2309.15728/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 27 Sep 2023.

## Built on by (most cited first, 7 of 7)

- [Bridging Theory and Practice in Link Representation with Graph Neural Networks](https://mlsanity.com/paper/2506.24018/) (2025)
- [Dynamic Neural Dowker Network: Approximating Persistent Homology in Dynamic Directed Graphs](https://mlsanity.com/paper/2408.09123/) (2024)
- [Haar-Laplacian for directed graphs](https://mlsanity.com/paper/2411.15527/) (2024)
- [LGAN: An Efficient High-Order Graph Neural Network via the Line Graph Aggregation](https://mlsanity.com/paper/2512.10735/) (2025)
- [LineMVGNN: Anti-Money Laundering with Line-Graph-Assisted Multi-View Graph Neural Networks](https://mlsanity.com/paper/2603.23584/) (2026)
- [Towards characterizing the value of edge embeddings in Graph Neural Networks](https://mlsanity.com/paper/2410.09867/) (2024)
- [GNNs Meet Sequence Models Along the Shortest-Path: an Expressive Method for Link Prediction](https://mlsanity.com/paper/2507.07138/) (2025)

## Abstract (arXiv)

In real-world networks, predicting the weight (strength) of links is as crucial as predicting the existence of the links themselves. Previous studies have primarily used shallow graph features for link weight prediction, limiting the prediction performance. In this paper, we propose a new link weight prediction method, namely Line Graph Neural Networks for Link Weight Prediction (LGLWP), which learns deeper graph features through deep learning. In our method, we first extract the enclosing subgraph around a target link and then employ a weighted graph labeling algorithm to label the subgraph nodes. Next, we transform the subgraph into the line graph and apply graph convolutional neural networks to learn the node embeddings in the line graph, which can represent the links in the original subgraph. Finally, the node embeddings are fed into a fully-connected neural network to predict the weight of the target link, treated as a regression problem. Our method directly learns link features, surpassing previous methods that splice node features for link weight prediction. Experimental results on six network datasets of various sizes and types demonstrate that our method outperforms state-of-the-art methods.