Disentangled Deep MHP

The implementation of our ICDM-21 paper "Disentangled Deep Multivariate Hawkes Process".

Requirements

python == 3.6.2

numpy == 1.17.4

scipy == 1.3.1

torch == 1.6.0

Dataset

Two commercial datasets are available here:

• ML-1M
• Taobao

The format of data is given as:

entity_id type_id timestamp

Two social datasets provided by previous works can be downloaded here:

• Retweets
• StackOverflow

Running the Code Command

For running ML-1M:

CUDA_VISIBLE_DEVICES=0 python -u train.py --model d2mhp --undebug --data_dir ml-1m --dis_k 4 --mi_loss 0.5

For running Taobao:

CUDA_VISIBLE_DEVICES=0 python -u train.py --model d2mhp --undebug --data_dir taobao_item --dis_k 6 --mi_loss 0.5

For running Retweets:

CUDA_VISIBLE_DEVICES=0 python -u train.py --model d2mhp --undebug --data_dir retweet --dis_k 2 --mi_loss 1

For running StackOverflow:

CUDA_VISIBLE_DEVICES=0 python -u train.py --model d2mhp --undebug --data_dir so --dis_k 4 --mi_loss 1