针对大语言模型(LLM)在零样本重排序任务中存在的标签语义理解不足、关系建模模糊和计算成本过高的问题,提出基于层次过滤与标签语义扩展的重排序方法HFLS(Hierarchical Filtering and Label Semantics)。该方法构建多级标签语义扩展路...针对大语言模型(LLM)在零样本重排序任务中存在的标签语义理解不足、关系建模模糊和计算成本过高的问题,提出基于层次过滤与标签语义扩展的重排序方法HFLS(Hierarchical Filtering and Label Semantics)。该方法构建多级标签语义扩展路径,并设计“关键词匹配→语义关联→领域知识整合”的递进式提示策略引导LLM实现深度相关性推理;同时,引入分层过滤机制,在降低计算复杂度的同时保留高潜力候选文档。实验结果表明:在TRECDL2019等7个基准数据集上,HFLS相较于Pointwise.qg、Pointwise.yes_no和Pointwise.3Label等Pointwise方法的NDCG@10(归一化折损累积增益)指标分别平均提升了21.92%、13.43%和8.59%;而在推理效率方面,HFLS的单个查询处理时延较Listwise方法、Pairwise方法和Setwise方法分别降低了91.06%、68.87%和33.54%。展开更多
针对现有知识推理补全方法在可解释性、计算效率和不可见实体预测三者间无法兼顾的问题,提出了基于中心路径结构的PGOCZ(Path Graph Of Central Zone)。PGOCZ利用实体局部结构上路径间的关联性和无关性,构建相互关联的路径结构作为可解...针对现有知识推理补全方法在可解释性、计算效率和不可见实体预测三者间无法兼顾的问题,提出了基于中心路径结构的PGOCZ(Path Graph Of Central Zone)。PGOCZ利用实体局部结构上路径间的关联性和无关性,构建相互关联的路径结构作为可解释性依据;通过关注实体局部路径间的关联与差异,避免重复子路径,提高计算效率;利用局部路径标识相对位置与语义信息,预测不可见实体。PGOCZ在归纳推理任务FB15k-237数据集和转导推理任务NELL-995数据集上表现最佳,且在多数据集上均能以不到30%的数据量达到稳定,有效兼顾了可解释性、计算效率和图谱外实体预测,验证了模型的有效性。展开更多
The performance of deep recommendation models degrades significantly under data poisoning attacks.While adversarial training methods such as Vulnerability-Aware Training(VAT)enhance robustness by injecting perturbatio...The performance of deep recommendation models degrades significantly under data poisoning attacks.While adversarial training methods such as Vulnerability-Aware Training(VAT)enhance robustness by injecting perturbations into embeddings,they remain limited by coarse-grained noise and a static defense strategy,leaving models susceptible to adaptive attacks.This study proposes a novel framework,Self-Purification Data Sanitization(SPD),which integrates vulnerability-aware adversarial training with dynamic label correction.Specifically,SPD first identifies high-risk users through a fragility scoring mechanism,then applies self-purification by replacing suspicious interactions with model-predicted high-confidence labels during training.This closed-loop process continuously sanitizes the training data and breaks the protection ceiling of conventional adversarial training.Experiments demonstrate that SPD significantly improves the robustness of both Matrix Factorization(MF)and LightGCN models against various poisoning attacks.We show that SPD effectively suppresses malicious gradient propagation and maintains recommendation accuracy.Evaluations on Gowalla and Yelp2018 confirmthat SPD-trainedmodels withstandmultiple attack strategies—including Random,Bandwagon,DP,and Rev attacks—while preserving performance.展开更多
文摘针对大语言模型(LLM)在零样本重排序任务中存在的标签语义理解不足、关系建模模糊和计算成本过高的问题,提出基于层次过滤与标签语义扩展的重排序方法HFLS(Hierarchical Filtering and Label Semantics)。该方法构建多级标签语义扩展路径,并设计“关键词匹配→语义关联→领域知识整合”的递进式提示策略引导LLM实现深度相关性推理;同时,引入分层过滤机制,在降低计算复杂度的同时保留高潜力候选文档。实验结果表明:在TRECDL2019等7个基准数据集上,HFLS相较于Pointwise.qg、Pointwise.yes_no和Pointwise.3Label等Pointwise方法的NDCG@10(归一化折损累积增益)指标分别平均提升了21.92%、13.43%和8.59%;而在推理效率方面,HFLS的单个查询处理时延较Listwise方法、Pairwise方法和Setwise方法分别降低了91.06%、68.87%和33.54%。
文摘针对现有知识推理补全方法在可解释性、计算效率和不可见实体预测三者间无法兼顾的问题,提出了基于中心路径结构的PGOCZ(Path Graph Of Central Zone)。PGOCZ利用实体局部结构上路径间的关联性和无关性,构建相互关联的路径结构作为可解释性依据;通过关注实体局部路径间的关联与差异,避免重复子路径,提高计算效率;利用局部路径标识相对位置与语义信息,预测不可见实体。PGOCZ在归纳推理任务FB15k-237数据集和转导推理任务NELL-995数据集上表现最佳,且在多数据集上均能以不到30%的数据量达到稳定,有效兼顾了可解释性、计算效率和图谱外实体预测,验证了模型的有效性。
文摘The performance of deep recommendation models degrades significantly under data poisoning attacks.While adversarial training methods such as Vulnerability-Aware Training(VAT)enhance robustness by injecting perturbations into embeddings,they remain limited by coarse-grained noise and a static defense strategy,leaving models susceptible to adaptive attacks.This study proposes a novel framework,Self-Purification Data Sanitization(SPD),which integrates vulnerability-aware adversarial training with dynamic label correction.Specifically,SPD first identifies high-risk users through a fragility scoring mechanism,then applies self-purification by replacing suspicious interactions with model-predicted high-confidence labels during training.This closed-loop process continuously sanitizes the training data and breaks the protection ceiling of conventional adversarial training.Experiments demonstrate that SPD significantly improves the robustness of both Matrix Factorization(MF)and LightGCN models against various poisoning attacks.We show that SPD effectively suppresses malicious gradient propagation and maintains recommendation accuracy.Evaluations on Gowalla and Yelp2018 confirmthat SPD-trainedmodels withstandmultiple attack strategies—including Random,Bandwagon,DP,and Rev attacks—while preserving performance.
