基于知识图谱的推荐方法是推荐系统的研究热点之一,利用用户历史行为及物品特征在知识图谱结构化表示数据的辅助下解决推荐系统数据稀疏性及冷启动问题。但是用户的兴趣易受特定物品所影响,而知识图谱以结构化形式存储数据,实体与实体...基于知识图谱的推荐方法是推荐系统的研究热点之一,利用用户历史行为及物品特征在知识图谱结构化表示数据的辅助下解决推荐系统数据稀疏性及冷启动问题。但是用户的兴趣易受特定物品所影响,而知识图谱以结构化形式存储数据,实体与实体之间仅存在链路关系,这就导致了单纯利用知识图谱的推荐方法在点击率预测方面性能较差。基于此,提出一种基于局部影响力与深层偏好传播的推荐方法(local influence and deep preference propagation,LIDP),充分利用知识图谱结构化数据在偏好传播中存在实体影响力的优点。LIDP模型首先对知识图谱逐层偏好传播获取数据影响力权重并根据所获数据权重进行局部影响力计算;其次对局部影响力进行用户历史行为的兴趣增强表示进而获取用户表示;最后对用户表示与物品的向量表示进行内积操作以获取最终交互概率。LIDP模型在MovieLens-1M数据集上相比最优基准模型GNRF,AUC、ACC、MAE和F 1值分别提高了0.16%、0.52%、0.87%、0.21%;在Book-Crossing数据集上,这些提升分别为0.45%、2.14%、1.29%、0.93%。实验结果表明,LIDP模型能有效获取深层次用户兴趣偏好,在推荐系统中具有良好的性能和效果,可以为用户提供更好的个性化推荐服务。展开更多
The effect of the local hard zone (LHZ) distributed in the coarse grained HAZ (CGHAZ) has been analyzed by 2-dimensional FEM on a mechanical model of the weld CGHAZ. The existence of the LHZ elevates considerably the ...The effect of the local hard zone (LHZ) distributed in the coarse grained HAZ (CGHAZ) has been analyzed by 2-dimensional FEM on a mechanical model of the weld CGHAZ. The existence of the LHZ elevates considerably the stress in LHZ and causes the discontinuity of strain at the boundary between the LHZ and the matrix. The stress distribution in the LHZ is strongly affected by the shape of the LHZ. In a slender LHZ almost the whole region in the LHZ is exposed to elevated stress, whereas in the massive LHZ only the edge region sustains high stress. The longer the LHZ becomes, the more the highly stressed area, and the peak stress in the LHZ grows even under the same volume fraction of the LHZ. These results indicate that the slender LHZ brings about unstable fracture at a lower load level than the blocky LHZ. This tendency was confirmed by CTOD test results on the weld CGHAZ of a high-strength steel. The CGHAZ with elongated M-A constituents fractures at apparently lower critical CTOD than the CGHAZ with massive M-A constituents. In conclusion, the control of the shape of the M-A constituent has a striking effect on the toughness improvement of CGHAZ.展开更多
文摘基于知识图谱的推荐方法是推荐系统的研究热点之一,利用用户历史行为及物品特征在知识图谱结构化表示数据的辅助下解决推荐系统数据稀疏性及冷启动问题。但是用户的兴趣易受特定物品所影响,而知识图谱以结构化形式存储数据,实体与实体之间仅存在链路关系,这就导致了单纯利用知识图谱的推荐方法在点击率预测方面性能较差。基于此,提出一种基于局部影响力与深层偏好传播的推荐方法(local influence and deep preference propagation,LIDP),充分利用知识图谱结构化数据在偏好传播中存在实体影响力的优点。LIDP模型首先对知识图谱逐层偏好传播获取数据影响力权重并根据所获数据权重进行局部影响力计算;其次对局部影响力进行用户历史行为的兴趣增强表示进而获取用户表示;最后对用户表示与物品的向量表示进行内积操作以获取最终交互概率。LIDP模型在MovieLens-1M数据集上相比最优基准模型GNRF,AUC、ACC、MAE和F 1值分别提高了0.16%、0.52%、0.87%、0.21%;在Book-Crossing数据集上,这些提升分别为0.45%、2.14%、1.29%、0.93%。实验结果表明,LIDP模型能有效获取深层次用户兴趣偏好,在推荐系统中具有良好的性能和效果,可以为用户提供更好的个性化推荐服务。
文摘The effect of the local hard zone (LHZ) distributed in the coarse grained HAZ (CGHAZ) has been analyzed by 2-dimensional FEM on a mechanical model of the weld CGHAZ. The existence of the LHZ elevates considerably the stress in LHZ and causes the discontinuity of strain at the boundary between the LHZ and the matrix. The stress distribution in the LHZ is strongly affected by the shape of the LHZ. In a slender LHZ almost the whole region in the LHZ is exposed to elevated stress, whereas in the massive LHZ only the edge region sustains high stress. The longer the LHZ becomes, the more the highly stressed area, and the peak stress in the LHZ grows even under the same volume fraction of the LHZ. These results indicate that the slender LHZ brings about unstable fracture at a lower load level than the blocky LHZ. This tendency was confirmed by CTOD test results on the weld CGHAZ of a high-strength steel. The CGHAZ with elongated M-A constituents fractures at apparently lower critical CTOD than the CGHAZ with massive M-A constituents. In conclusion, the control of the shape of the M-A constituent has a striking effect on the toughness improvement of CGHAZ.
