feiyang / metaembedding

Codes for our SIGIR-2019 paper "Warm Up Cold-start Advertisements: Improving CTR Predictions via Learning to Learn ID Embeddings"

MetaEmbedding

Codes for our SIGIR-2019 paper:

Warm Up Cold-start Advertisements: Improving CTR Predictions via Learning to Learn ID Embeddings

This repo includes an example for training Meta-Embedding upon a deepFM model on the binarized MovieLens-1M dataset. The dataset is preprocessed and splitted already.

Requirements: Python 3 and TensorFlow.

Bibtex

@inproceedings{pan2019warm,
 author = {Pan, Feiyang and Li, Shuokai and Ao, Xiang and Tang, Pingzhong and He, Qing},
 title = {Warm Up Cold-start Advertisements: Improving CTR Predictions via Learning to Learn ID Embeddings},
 booktitle = {Proceedings of the 42Nd International ACM SIGIR Conference on Research and Development in Information Retrieval},
 series = {SIGIR'19},
 year = {2019},
 isbn = {978-1-4503-6172-9},
 location = {Paris, France},
 pages = {695--704},
 numpages = {10},
 url = {http://doi.acm.org/10.1145/3331184.3331268},
 doi = {10.1145/3331184.3331268},
 acmid = {3331268},
 publisher = {ACM},
 address = {New York, NY, USA},
} 

metaembedding's Issues

Hi，想问下论文里提到的Online Training是怎么个流程呢

论文里提到的“Note that Meta-Embedding not only can be trained with offline data set, but also can be trained online with minor modifications by using the emerging new IDs as the training examples.”
如果想在近实时的kafka样本上进行训练Meta网络，应该怎么个“minor modifications”呢，有时间的话还不望赐教。

在论文里为什么只训一个epoch， 而不是训到收敛呢？

您好，感谢开源；
我正在做冷启动相关的工作，想请教您为什么实验设置成只train 一个epoch，正常冷启动的场景应该是训练到收敛？您有尝试过训练到收敛之后的结果吗？
thx

ME_loss 这个变量没用到？

是这个地方应该是Me_loss?

In the data, the Three pickle files are the same(test_oneshot_a(b,c).pkl)