Prefix Alignment with Temporal History (PATH), which aggregates transition statistics from the training corpus over recent interactions with exponential decay to construct history-conditioned two-token prefix targets, is proposed, showing consistent improvements in recommendation performance and higher full-SID surviva...
Hong-Liang Sun, Lian-Jie Li, Bolin Zhang et al.· 1 citation
Experiments on LoCoMo and MemConflict demonstrate that the proposed EMIR framework and intent-guided multi-round retrieval mechanism improves long-term memory utilization, dynamic and static conflict handling, and complex retrieval performance, achieving relative improvements of more than 12\% in certain categories.
Jin-Lan Liu, Hong-Liang Sun, Yong Wang et al.· 0 citations
GARDRec, a Graph-grounded Adaptive Reasoning and Decision-aware Recommendation framework for LLM-based next-item ranking, is proposed, which improves candidate-ranking performance over representative baselines and verifies the contributions of graph projection, neighborhood retrieval, explicit decision features, rankin...
Yong Wang, Hong-Liang Sun, Jin-Lan Liu et al.· 0 citations
Temporal Autoregressive Alignment (TAAL) is proposed, which improves NDCG@10 over the standard baseline by 39.5%, and aligns the early-prefix distribution with a forward KL objective during training and during inference, it calibrates candidate scores with pointwise mutual information (PMI) to reduce the influence of g...
Lian-Jie Li, Zhi-Ying Tu, Dianhui Chu et al.· 0 citations
Experiments show that DuPLeR achieves robust performance in data-scarce KGC scenarios, and a dual-pathway multimodal enhancement module regulates message passing with query-relevant multimodal signals and supplements entity representations after graph propagation.
