Freqrec Eval

---
name: freqrec-eval
description: Evaluates a sequential recommendation model's ability to predict the next item in a user's interaction history by jointly modeling intra-session and inter-session behavioral dynamics. It probes the model's recommendation accuracy, robustness to noisy cross-domain data, and stability under sparse interaction conditions. Use when the user wants to benchmark on Amazon Beauty, Sports & Outdoors, Toys & Games, or asks about evaluating this task. Reports HR@K, NDCG@K.
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2511.06285
  bibtex_key: he2025exploiting
  confidence: high
---

# freqrec-eval

> Exploiting Inter-Session Information with Frequency-enhanced Dual-Path Networks for Sequential Recommendation — Peng He et al. (arXiv:2511.06285, 2025)

## What this evaluates

Evaluates a sequential recommendation model's ability to predict the next item in a user's interaction history by jointly modeling intra-session and inter-session behavioral dynamics. It probes the model's recommendation accuracy, robustness to noisy cross-domain data, and stability under sparse interaction conditions.

## Datasets

- **Amazon Beauty** — total ?; splits: train (-1), val (-1), test (-1)
- **Sports & Outdoors** — total ?; splits: train (-1), val (-1), test (-1)
- **Toys & Games** — total ?; splits: train (-1), val (-1), test (-1)

## Metrics

- `HR@K` **(primary)** — range: [0, 1]
  - Binary indicator: 1 if the ground-truth next item appears in the top-K predicted items, 0 otherwise.
- `NDCG@K` **(primary)** — range: [0, 1]
  - Normalized Discounted Cumulative Gain for a single relevant item: 1/log2(2) if the ground-truth item is in the top-K, 0 otherwise.

## Input / output format

**Input**: Ordered sequence of user-item interactions (sessions) representing historical behavior.

**Output**: Top-K ranked list of candidate items for the next interaction.

## Scoring recipe

```python
def compute_hr_ndcg(preds, gold, k):
    hit = 1.0 if gold in preds[:k] else 0.0
    dcg = hit / math.log2(2)
    idcg = 1.0 / math.log2(2)
    ndcg = dcg / idcg
    return hit, ndcg
```

## Common pitfalls

- Data preprocessing and train/val/test splits are borrowed from prior work and not explicitly detailed in the main text.
- K values vary by experiment: K=10/20 for main benchmarks, but K=5 is used for sparsity and noise-resilience evaluations.
- Noise evaluation uses a cross-domain protocol (train on aggregate of Automotive, CDs, Grocery; test on each independently) rather than standard in-domain testing.

## Evidence (verbatim from paper)

> To comprehensively evaluate the model’s recommendation accuracy, we employ the top-$K$ evaluation method commonly used in SR, including Hit Rate (HR@$K$) and Normalized Discounted Cumulative Gain (NDCG@$K$), with $K$ set at 10 and 20.

## Citation

```bibtex
@misc{he2025exploiting,
  title={Exploiting Inter-Session Information with Frequency-enhanced Dual-Path Networks for Sequential Recommendation},
  author={Peng He et al.},
  year={2025},
  note={arXiv:2511.06285}
}
```

- arXiv: 2511.06285