Decision-Making Context Interaction Network for Click-Through Rate Prediction

Authors: Xiang Li, Shuwei Chen, Jian Dong, Jin Zhang, Yongkang Wang, Xingxing Wang, Dong Wang

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In the experiments on public and industrial datasets, DCIN significantly outperforms the state-of-the-art methods. Notably, the model has obtained the improvement of CTR+2.9%/CPM+2.1%/GMV+1.5% for online A/B testing and served the main traffic of Meituan Waimai advertising system.
Researcher Affiliation Industry Xiang Li*, Shuwei Chen*, Jian Dong, Jin Zhang, Yongkang Wang, Xingxing Wang, Dong Wang Meituan, Beijing, China {lixiang172,chenshuwei04,dongjian03,zhangjin11}@meituan.com {wangyongkang03,wangxingxing04,wangdong07}@meituan.com
Pseudocode No The paper describes its methods using mathematical equations and textual explanations, but it does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statement about releasing open-source code for the described methodology, nor does it include a link to a code repository.
Open Datasets Yes Since few large scale datasets contain both behavior page information and pre-ranking candidates, we construct a dataset based on the publicly available Avito1 dataset. [...] 1https://www.kaggle.com/c/avito-context-ad-clicks/data.
Dataset Splits Yes We use the data from 2015-0428 to 2015-05-18 as the training set, 2015-05-19 as the validation set and 2015-05-20 as the testing set. [...] we collected the real behavior pages exposed to the users and the corresponding candidates from the online service logs of Meituan Waimai App from 2022-05-25 to 2022-06-10 as the training set, and collected the data in 2022-06-11 as the validation set and 2022-06-12 as the testing set.
Hardware Specification No The paper states: 'Our models are trained in a large-scale machine learning platform in Meituan.' This is a general description of the computing environment but does not provide specific hardware details such as GPU/CPU models or memory.
Software Dependencies No The paper mentions using 'Ada Grad (Duchi, Hazan, and Singer 2011) to optimize all the networks' but does not provide specific version numbers for any software dependencies (e.g., programming languages, deep learning frameworks, or libraries).
Experiment Setup Yes The final MLPs in all experiments contain two layers with 256 and 128 hidden units. We use Ada Grad (Duchi, Hazan, and Singer 2011) to optimize all the networks. The hyper-parameters are set as follows: for the constructed Avito dataset, the length of click sequence S = 5, the number of intra-page items P = 5, and the number of candidate items C = 20; for Meituan Ads dataset, we set the length of click sequence S = 20, the number of intra-page items P = 10, and the number of candidate items C = 60. The values of k in explicit/implicit CIUs are selected experimentally (see ablation study for details).