A Multimodal Approach for Improving Market Price Estimation in Online Advertising

This is a repository of experiment code supporting A Multimodal Approach for Improving Market Price Estimation in Online Advertising.

For any problems, please report the issues here.

Quirk Start

Prepare Dataset

Before run the demo, please first check the GitHub project make iPinYou data for pre-processing the iPinYou dataset. Or you can download the processed dataset from this link with extracting code h12c.

Then, please create a folder named dataset, and put the dataset in it. The file tree looks like this:

GMM
│───README.md
│
└───dataset
│   └───make-ipinyou-data
│       │   1458
│       │   2259
│       │   ...
...

Encode Dataset

Please run the following code to encode the dataset

cd ./shell
bash ./ipinyou_dataset_encode.sh

You can find the running logs in this directory /result/$advertiser/log/dataset_encode

Run GMM and CGMM

Please run the following code to train and evaluate the GMM and CGMM

bash ./GMM.sh
bash ./CGMM.sh

You can find the running logs in this directories /result/$advertiser/log/GMM and /result/$advertiser/log/CGMM

Run Baselines

Please run the following code to train and evaluate the baselines

bash ./tanh_LT_LG.sh
bash ./CLR.sh
bash ./SM.sh
bash ./DCL.sh
bash ./DLF.sh

You can find the running logs in these directories /result/$advertiser/log/tanh_LT_LG, /result/$advertiser/log/CLR, /result/$advertiser/log/SM, /result/$advertiser/log/DCL and /result/$advertiser/log/DLF.

The code of DCL and DLF is forked from this repository.