The relevant datasets and model codes of this paper are summarized as follows.
To validate the performance of BASM, a takeaway industrial dataset Ele.me and a public spatiotemporal recommendation dataset. The experimental data set statistics are shown in Table I below:
Table I: The basic Statistics of datasets.
| Datasets | Total Size | Feature | Users | Items | Clicks | Mean Length of User Behaviors |
|---|---|---|---|---|---|---|
| Ele.me (Industry) | 2380427866 | 417 | 81086293 | 547354 | 86735276 | 42.86 |
| Spatiotemporal Public Data | 177114244 | 38 | 14427689 | 7446116 | 3140831 | 41.19 |
Ele.me is collected from the industrial recommendation platform. It has more than 80 million users and 2 billion samples. 45-day samples are used for training and the samples of the following day are used for testing.
The dataset contains a total of more than 170 million pieces of data over 8 days, including 7 days of data for model training and 1 day of data for testing. The dataset can be downloaded from link https://tianchi.aliyun.com/dataset/dataDetail?dataId=131047
Since the model is currently deployed in the Ele.me commercial food delivery system, we will only open source all publicly available comparison models in the paper (including Wide & Deep, DIN, AutoInt), which can be reproduced offline through DeepCTR.
The comparison models are shown in the table below:
| Model | Paper |
|---|---|
| Wide & Deep | [DLRS 2016]Wide & Deep Learning for Recommender Systems |
| Deep Interest Network | [KDD 2018]Deep Interest Network for Click-Through Rate Prediction |
| AutoInt | [CIKM 2019]AutoInt: Automatic Feature Interaction Learning via Self-Attentive Neural Networks |
| M2M | [WSDM 2022]Leaving No One Behind: {A} Multi-Scenario Multi-Task Meta Learning Approach for Advertiser Modeling |
| STAR | [CIKM 2021]One model to serve all: Star topology adaptive recommender for multi-domain ctr prediction |
| APG | [NeurIPS 2022]APG: Adaptive Parameter Generation Network for Click-Through Rate Prediction |
The source code of the comparison models can be seen in the models folder of this project.