Recommendation-Improved

推荐系统---实验+复现+创新

书籍阅读复现阶段

• 推荐系统实践 https://item.jd.com/11007625.html

论文阅读复现阶段

• DeepCoNN https://arxiv.org/abs/1701.04783 Done
• NRPA https://arxiv.org/abs/1905.12480v1 Done
• DAML https://dl.acm.org/doi/10.1145/3292500.3330906 Done

自己的论文

• DAML-Improved 对DAML模型在id信息融合方式上借鉴NRPA模型的思想 Done (模型见/pic/IDAML.png)
• DAML-Distance 对DAML模型在id信息融合方式上借鉴NRPA模型的思想并对距离公式进行调研 Done
  • 欧式距离（常规 + 标准化加权）
  • 皮尔逊相关系数

数据集

• MOVIE LINES (http://www.grouplens.org/node/73)
• Amazon 5-core(https://nijianmo.github.io/amazon/index.html)
• YELP (https://www.yelp.com/dataset/download)

参数(reviews_Sports_and_Outdoors_5 Best)

⚠️DropOut概率参数本仓库全部置为了1，无参考价值，为Demo，请具体根据实际场景进行调整

#DeepCoNN
BATCH_SIZE          = 64
EPOCHS              = 50
LEARNING_RATE       = 0.02
CONV_LENGTH         = 3
CONV_KERNEL_NUM     = 32
FM_K                = 1 #Factorization Machine 交叉向量维度
LATENT_FACTOR_NUM   = 64
GPU_DEVICES         = 0

#NRPA
BATCH_SIZE          = 128
EPOCHS              = 50
LEARNING_RATE       = 0.01
CONV_LENGTH         = 3
CONV_KERNEL_NUM     = 28
FM_K                = 1 #Factorization Machine 交叉向量维度
LATENT_FACTOR_NUM   = 56
GPU_DEVICES         = 0
ID_EMBEDDING_DIM    = 32
ATTEN_VEC_DIM       = 32

#DAML
BATCH_SIZE          = 128
EPOCHS              = 50
LEARNING_RATE       = 0.001
CONV_LENGTH         = 3
CONV_KERNEL_NUM     = 16
FM_K                = 1 #Factorization Machine 交叉向量维度
LATENT_FACTOR_NUM   = 58
GPU_DEVICES         = 0
ID_EMBEDDING_DIM    = 32
ATTEN_VEC_DIM       = 16
ATT_CONV_SIZE       = 3

#ImprovedDAML
BATCH_SIZE          = 24
EPOCHS              = 75
LEARNING_RATE       = 0.001
CONV_LENGTH         = 3
CONV_KERNEL_NUM     = 16
FM_K                = 1 #Factorization Machine 交叉向量维度
LATENT_FACTOR_NUM   = 32
GPU_DEVICES         = 0
ID_EMBEDDING_DIM    = 32
ATTEN_VEC_DIM       = 16
REVIEW_SIZE         = 15
ATT_CONV_SIZE       = 3

#DistanceImprovedDAML (Standardized Euclidean distance )
BATCH_SIZE          = 24
EPOCHS              = 75
LEARNING_RATE       = 0.001
CONV_LENGTH         = 3
CONV_KERNEL_NUM     = 16
FM_K                = 1 #Factorization Machine 交叉向量维度
LATENT_FACTOR_NUM   = 32
GPU_DEVICES         = 0
ID_EMBEDDING_DIM    = 32
ATTEN_VEC_DIM       = 16
REVIEW_SIZE         = 15
ATT_CONV_SIZE       = 3

训练结果展示

test文件夹中的json文件

命名格式:

train_{model_name}_{dataset_name}_{reviews_length}_{reviews_size}_{user_num}_{item_num}

环境

• python3.7
• pytorch => torch 1.0.0 && torchvision 0.2.1
• gensim => gensim 3.8.1
• numpy => numpy 1.16.0
• pandas => pandas 0.25.3
• tqdm => tqdm 4.42.0
• py2neo => py2neo 4.3.0