ELM

A Evaluation-based Large Model for Multi-step Behavior Prediction in Intelligent Consumer Electronics

Overview of ELM

This repository includes the source code of the paper accepted by xxxxxx.

Dependencies

• Dependencies can be installed using requirements.txt.

Dataset:

• We use NYT, icew14, icew15 and finance dataset for event prediction.

Pretrained Checkpoint:

The 1-step prediction results are:

Dataset	NYT	ICEWS14	ICEWS15	Finance
INSEP	81.37%	81.52%	86.75%	79.15%

The 2-step prediction results are:

Dataset	H@1	H@3	H@10
NYT	72.56%	93.90%	97.27%
ICEWS14	73.31%	93.64%	95.55%
ICEWS15	79.53%	91.37%	96.07%
Finance	70.21%	91.56%	95.18%

The 3-step prediction results are:

Dataset	H@1	H@3	H@10
NYT	62.60%	88.98%	96.01%
ICEWS14	62.17%	88.33%	96.45%
ICEWS15	67.54%	85.13%	95.52%
Finance	60.42%	87.13%	95.54%

The 4-step prediction results are:

Dataset	H@1	H@3	H@10
NYT	53.09%	84.16%	94.89%
ICEWS14	52.94%	85.63%	94.76%
ICEWS15	52.77%	87.27%	95.37%
Finance	53.55%	84.76%	94.54%

Training and testing:

• Install all the requirements from ./requirements.txt.

• Commands for reproducing the reported results:

  (1) Evaluator

  python Evaluator.py

• epoch=20

• lamda=15

• skip_n_val_epoch:train_model->1000; test_model->0

  (2) Generator

  python Generator.py

• epoch=60

• lamda=15

• skip_n_val_epoch:train_model->1000; test_model->0

Citation

If you used our work or found it helpful, please use the following citation:

@inproceedings{ELM,
    title = "A Evaluation-based Large Model for Multi-step Behavior Prediction in Intelligent Consumer Electronics",
    author = "Li, Jinpeng and
      Yu, Hang  and
      Wei, Subo 
    booktitle = "xxxxxx",
}