为缓解推荐系统中数据稀疏性问题,利用卷积神经网络CNN具有较强捕捉局部特征能力的优势,通过加入一个调节层,提出一种改进CNN的局部相似性预测推荐模型LSPCNN。新模型对初始用户-项目评分矩阵进行迭代调整,使用户兴趣偏好局部特征化,再...为缓解推荐系统中数据稀疏性问题,利用卷积神经网络CNN具有较强捕捉局部特征能力的优势,通过加入一个调节层,提出一种改进CNN的局部相似性预测推荐模型LSPCNN。新模型对初始用户-项目评分矩阵进行迭代调整,使用户兴趣偏好局部特征化,再融合CNN对缺失评分进行预测,从而实施个性化推荐。实验结果表明,LSPCNN模型在不同数据稀疏度下的 MAE 值较传统推荐方法平均下降4%,有效缓解了数据稀疏性,提高了推荐系统的性能。展开更多
To combine the technical functions and advantages of solid-core fiber Bragg gratings(FBGs) and hollow-core optical fibers(HCFs), the hollow and filled FBGs in nano-bore optical fibers(NBFs) with nano-bore in the GeO2-...To combine the technical functions and advantages of solid-core fiber Bragg gratings(FBGs) and hollow-core optical fibers(HCFs), the hollow and filled FBGs in nano-bore optical fibers(NBFs) with nano-bore in the GeO2-doped core are proposed.The fundamental mode field, effective mode index, and confinement loss of NBF with 50 nm–7 μm-diameter hollow and filled nano-bore are numerically investigated by the finite element method.The reflected spectra of FBGs in NBFs are obtained by the transmission matrix method.The hollow FBGs in NBFs can be acheived with ~5% power fraction in the bore and the ~0.9 reflectivity when bore diameter is less than 3 μm.The filled FBGs can be realized with^1% power fraction and 0.98 reflectivity with different fillings including o-xylene, trichloroethylene, and chloroform for 800-nm bore diameter.The feasibility of the index sensing by our proposed NBF FBG is also analyzed and discussed.The experimental fabrication of hollow and filled FBGs are discussed and can be achieved by current techniques.The aim of this work is to establish a principle prototype for investigating the HCFs and solid-core FBGs-based fiber-optic platforms,which are useful for applications such as the simultaneous chemical and physical sensing at the same position.展开更多
文摘为缓解推荐系统中数据稀疏性问题,利用卷积神经网络CNN具有较强捕捉局部特征能力的优势,通过加入一个调节层,提出一种改进CNN的局部相似性预测推荐模型LSPCNN。新模型对初始用户-项目评分矩阵进行迭代调整,使用户兴趣偏好局部特征化,再融合CNN对缺失评分进行预测,从而实施个性化推荐。实验结果表明,LSPCNN模型在不同数据稀疏度下的 MAE 值较传统推荐方法平均下降4%,有效缓解了数据稀疏性,提高了推荐系统的性能。
基金Project supported by the Beijing Natural Science Foundation,China(Grant No.4192047)the Fundamental Research Funds for the Central Universities,China(Grant Nos.2018JBM070 and 2018JBM065)the National Natural Science Foundation of China(Grant No.61675019)
文摘To combine the technical functions and advantages of solid-core fiber Bragg gratings(FBGs) and hollow-core optical fibers(HCFs), the hollow and filled FBGs in nano-bore optical fibers(NBFs) with nano-bore in the GeO2-doped core are proposed.The fundamental mode field, effective mode index, and confinement loss of NBF with 50 nm–7 μm-diameter hollow and filled nano-bore are numerically investigated by the finite element method.The reflected spectra of FBGs in NBFs are obtained by the transmission matrix method.The hollow FBGs in NBFs can be acheived with ~5% power fraction in the bore and the ~0.9 reflectivity when bore diameter is less than 3 μm.The filled FBGs can be realized with^1% power fraction and 0.98 reflectivity with different fillings including o-xylene, trichloroethylene, and chloroform for 800-nm bore diameter.The feasibility of the index sensing by our proposed NBF FBG is also analyzed and discussed.The experimental fabrication of hollow and filled FBGs are discussed and can be achieved by current techniques.The aim of this work is to establish a principle prototype for investigating the HCFs and solid-core FBGs-based fiber-optic platforms,which are useful for applications such as the simultaneous chemical and physical sensing at the same position.