Hierarchical Intention Embedding Network

This is our Tensorflow implementation for the paper:

Zuowu Zheng, Changwang Zhang, Xiaofeng Gao, and Guihai Chen. 2022. HIEN: Hierarchical Intention Embedding Network for Click-Through Rate Prediction. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’22), July 11–15, 2022, Madrid, Spain.

Environment Requirement

The code has been tested running under Python 3.5. The required packages are as follows:

• tensorflow == 1.4
• numpy == 1.14.3
• scipy == 1.1.0
• sklearn == 0.19.1
• pandas == 0.22.0 Note that we refer to the implementation of NGCF.

Dataset

1. Alimama Data and Tmall Data
2. Extract the files into the data/raw_data directory
3. Follow the code of the data preprocessing in DSIN to preprocess data
4. Then we need to extract structure information in item attributes. Take (adgroup_id, campaign_id, customer) as example, we build a dictionary for (campaign_id: adgroup_id) and (customer: campaign_id) respectively, which contains the relation dependencies of these attributes.

Example to Run the Codes

python main.py --regs [1e-5]
               --embed_size 128
               --lr 0.001
               --save_flag 0
               --pretrain 0
               --batch_size 4096
               --epoch 50
               --verbose 50
               --node_dropout [0.1]
               --mess_dropout [0.1,0.1,0.1]
               --tree_type gcn

Some important arguments:

• alg_type

  • It specifies the type of graph convolutional layer.
  • Here we provide three options:
    • ngcf , proposed in Neural Graph Collaborative Filtering, SIGIR2019. Usage: --alg_type ngcf.
    • gcn, proposed in Semi-Supervised Classification with Graph Convolutional Networks, ICLR2018. Usage: --alg_type gcn.
    • gcmc, propsed in Graph Convolutional Matrix Completion, KDD2018. Usage: --alg_type gcmc.

• adj_type

  • It specifies the type of laplacian matrix where each entry defines the decay factor between two connected nodes.
  • Here we provide four options:
    • ngcf (by default), where each decay factor between two connected nodes is set as 1(out degree of the node), while each node is also assigned with 1 for self-connections. Usage: --adj_type ngcf.
    • plain, where each decay factor between two connected nodes is set as 1. No self-connections are considered. Usage: --adj_type plain.
    • norm, where each decay factor bewteen two connected nodes is set as 1/(out degree of the node + self-conncetion). Usage: --adj_type norm.
    • gcmc, where each decay factor between two connected nodes is set as 1/(out degree of the node). No self-connections are considered. Usage: --adj_type gcmc.

• node_dropout

  • It indicates the node dropout ratio, which randomly blocks a particular node and discard all its outgoing messages. Usage: --node_dropout [0.1] --node_dropout_flag 1
  • Note that the arguement node_dropout_flag also needs to be set as 1, since the node dropout could lead to higher computational cost compared to message dropout.

• mess_dropout

  • It indicates the message dropout ratio, which randomly drops out the outgoing messages. Usage --mess_dropout [0.1,0.1,0.1].

• tree_type

  • It indicates the aggregator we used in attribute graph aggregation
  • Here we provide four options:
    • GCN
    • NGCF
    • LightGCN
    • Concat & Product (CP)