Path-Based Gradient Boosting for Graph-Level Prediction
Authors: Claudio Meggio, Johan Pensar, Riccardo De Bin
Summary
PathBoost is a gradient boosting method for predicting properties of entire graphs (like molecules or social networks) by learning discriminative path-based features directly from graph structure. It extends previous work by handling binary classification, incorporating multiple node/edge attributes through prefix decomposition, and automatically selecting anchor nodes (starting points for paths) based on attribute diversity. The method matches or beats graph neural networks and graph kernel methods on several benchmarks, especially for graphs with many nodes.
Main takeaways:
- Uses gradient tree boosting to learn which paths through a graph are predictive of the target property
- Automatically picks anchor nodes based on how diverse their categorical attributes are, removing the need for manual specification
- Handles multiple node and edge attributes by decomposing them into prefix-based features
- Competitive with or better than graph neural networks on half of benchmark datasets, particularly strong on larger graphs
- Offers an interpretable alternative to black-box graph neural networks by explicitly identifying discriminative path patterns
Relevance
Not directly related to my persona/midtraining work — this is about learning from graph-structured data (molecules, networks) rather than language model behavior or internal representations.
Threat model
Potential threat/caveat for clean result "Fine-tuning one persona on a two-marker chunk and another on the start marker plants the end marker at every donor answer's end, not chained to the start (LOW confidence)": this item discusses benchmark.
Abstract
arXiv:2605.08102v1 Announce Type: new Abstract: We propose PathBoost, a gradient tree boosting method for graph-level classification and regression that learns discriminative path-based features directly from the input graph structure. Building on a previous work, which was tailored to a specific chemistry application, PathBoost introduces three key extensions: (i) adaptation to binary classification through gradient boosting with a logistic loss, (ii) incorporation of multiple node and edge attributes into the path feature space via a prefix-based decomposition, and (iii) automatic anchor node selection based on categorical attribute diversity, eliminating the need for the user to specify the starting point of the considered path features. We compared PathBoost to graph neural networks and graph kernel approaches on several benchmark datasets, obtaining better results in half of them, and comparable results in the rest. PathBoost shows better performances on graphs with larger average node counts. Overall, the results demonstrate that path-based boosting methods can be competitive with more complex black-box approaches.