Research map / Label-efficient and robust learning / Recognition and long-tail

Augmenting Sequential Recommendation with Balanced Relevance and Diversity

AAAI 2025 · Label-efficient and robust learning › Recognition and long-tail · approach: Recommendation representation learning · 20 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes BASRec, a model-agnostic mixup-based data augmentation plugin for sequential recommendation that balances relevance and diversity via single-sequence and cross-sequence representation-level augmentation.

Problem

Data augmentation for sequential recommendation struggles to balance relevance (avoiding semantic drift) and diversity (enough variation) in augmented data, limiting performance gains.

Approach

Generate augmented samples in representation space: single-sequence operators (M-Reorder, M-Substitute) with uniformly sampled rate, mixed with the original representations and adaptively reweighted loss; plus cross-sequence nonlinear mixup (item-wise and feature-wise) between user representations in a batch, applied in a two-stage training schedule.

Key results

• BASRec improves SASRec by 33.84% on average and GRU4Rec by 72.04%, FMLP-Rec by 68.50% (average across datasets, per the paper).
• On SASRec, BASRec reaches N@10 0.0455 on Beauty vs 0.0338 base and 0.0372 for the best training-based baseline DiffuASR.
• Cosine similarity analysis: BASRec average similarity 0.9126 versus 0.9530 for ASReP and 0.9599 for CL4SRec, and 0.6539 for CMRSI, claimed to balance relevance and diversity.

Measured effects reported in the paper

Surprising findings

• Some heuristic augmentation methods (e.g., CMR, Ran, SW) degrade performance on some dataset/backbone combinations, which the authors attribute to noise in augmented data.
• CL4SRec achieves the best NDCG@10 on Yelp (0.0176) and HR@10 on Yelp (0.0355), outperforming BASRec on those metrics.

Limitations

• Balancing hyperparameters (alpha, a, b) must be tuned per dataset and the paper notes adaptive operator rates as future work.
• Paper only evaluates on SR backbones that are sequence encoders; generalization to other recommendation models is left for future work.
• Relevance/diversity measured only via cosine similarity of model output representations, an indirect proxy.

Experimental setup

Datasets: Beauty Sports Yelp Home

Benchmarks: Beauty Sports Yelp Home

Metrics: NDCG@10 HR@10 NDCG@20 HR@20

Baselines: GRU4Rec NextItNet SASRec FMLP-Rec Ran Ran-S SW CMR CMRSI ASReP DiffuASR CL4SRec

Builds on: mixup CL4SRec CMRSI DiffuASR ASReP

Closest papers by meaning

• Future Sight and Tough Fights: Revolutionizing Sequential Recommendation with FENRec (AAAI 2025 · similarity 0.90)
• SLMRec: Distilling Large Language Models into Small for Sequential Recommendation (ICLR 2025 · similarity 0.88)
• LLM4RSR: Large Language Models as Data Correctors for Robust Sequential Recommendation (AAAI 2025 · similarity 0.88)
• Customizing Language Models with Instance-wise LoRA for Sequential Recommendation (NeurIPS 2024 · similarity 0.88)
• LLM-ESR: Large Language Models Enhancement for Long-tailed Sequential Recommendation (NeurIPS 2024 · similarity 0.87)
• SRA-CL: Semantic Retrieval Augmented Contrastive Learning for Sequential Recommendation (NeurIPS 2025 · similarity 0.87)