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.

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:

The source code of the comparison models can be seen in the models folder of this project.