Hoprank Fewshot Node Class Eval

---
name: hoprank-fewshot-node-class-eval
description: Evaluates few-shot node classification on text-attributed graphs using self-supervised preference tuning. It probes the model's ability to leverage graph topology and anchor labels at inference without any supervised training, measuring both classification accuracy and inference efficiency. Use when the user wants to benchmark on Cora, Citeseer, Pubmed, or asks about evaluating this task. Reports Accuracy (%).
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2604.17271
  bibtex_key: wang2026hoprank
  confidence: high
---

# hoprank-fewshot-node-class-eval

> HopRank: Self-Supervised LLM Preference-Tuning on Graphs for Few-Shot Node Classification — Wang et al. (2026) (arXiv:2604.17271, 2026)

## What this evaluates

Evaluates few-shot node classification on text-attributed graphs using self-supervised preference tuning. It probes the model's ability to leverage graph topology and anchor labels at inference without any supervised training, measuring both classification accuracy and inference efficiency.

## Datasets

- **Cora, Citeseer, Pubmed** — total ?; splits: test (-1)

## Metrics

- `Accuracy (%)` **(primary)** — range: percent
  - Standard classification accuracy: the fraction of correctly predicted node labels out of the total number of nodes in the evaluation set.

## Input / output format

**Input**: Text-attributed graph nodes. At inference, K labeled nodes per class are provided as anchors. The model receives node text features and graph topology.

**Output**: Predicted class label for each node.

## Scoring recipe

```python
def compute_accuracy(predictions, gold_labels):
    correct = sum(1 for p, g in zip(predictions, gold_labels) if p == g)
    return (correct / len(gold_labels)) * 100
```

## Common pitfalls

- Using labeled nodes during the self-supervised training phase; HopRank strictly uses labels only at inference as anchors.
- Reporting single-run results without averaging over 5 different random seeds, as the protocol requires standard deviations.
- Ignoring the adaptive early-exit strategy when evaluating inference efficiency, which significantly reduces compute cost.

## Evidence (verbatim from paper)

> Table 4.1 summarizes the overall comparison. HopRank achieves the best accuracy across all few-shot settings on Citeseer and Pubmed, and remains competitive on Cora, despite using zero labeled nodes during training. All results are averaged over 5 runs with different random seeds; standard deviations are reported as subscripts.

## Citation

```bibtex
@misc{wang2026hoprank,
  title={HopRank: Self-Supervised LLM Preference-Tuning on Graphs for Few-Shot Node Classification},
  author={Wang et al. (2026)},
  year={2026},
  note={arXiv:2604.17271}
}
```

- arXiv: 2604.17271