Skip to content
/ LongCTR Public

LongCTR: A Long Sequence Modeling Benchmark for CTR Prediction

License

Notifications You must be signed in to change notification settings

reczoo/LongCTR

Repository files navigation

LongCTR

A Long Sequence Modeling Benchmark for CTR Prediction

Datasets

Models

No Model Publication
1 DCNv2 DCN V2: Improved Deep & Cross Network and Practical Lessons for Web-scale Learning to Rank Systems, WWW 2021.
2 FinalMLP FinalMLP: An Enhanced Two-Stream MLP Model for CTR Prediction, AAAI 2023.
3 DIN Deep Interest Network for Click-Through Rate Prediction, KDD 2018.
4 DIEN Deep Interest Evolution Network for Click-Through Rate Prediction, AAAI 2019.
5 TransAct TransAct: Transformer-based Realtime User Action Model for Recommendation at Pinterest, in KDD 2023.
6 SIM Search-based User Interest Modeling with Lifelong Sequential Behavior Data for Click-Through Rate Prediction, in CIKM 2020.
7 ETA Efficient Long Sequential User Data Modeling for Click-Through Rate Prediction, in DLP-KDD 2022.
8 SDIM Sampling Is All You Need on Modeling Long-Term User Behaviors for CTR Prediction, in CIKM 2022.
9 TWIN TWIN: TWo-stage Interest Network for Lifelong User Behavior Modeling in CTR Prediction at Kuaishou, in KDD 2023.
10 MIRRN Multi-granularity Interest Retrieval and Refinement Network for Long-Term User Behavior Modeling in CTR Prediction, in KDD 2025.

Benchmarking Results

In the following, we present benchmarking results for long sequence models across various sequence lengths in CTR prediction. Specifically, we establish the DIN model with a sequence length of 50 as the global baseline and report the relative improvements of other models compared to this baseline. Each row illustrates the performance variation of a specific model as the sequence length increases. Each column compares the performance of different models at a given sequence length, with the best performance in each column highlighted by underlining and the global best results highlighted in bold.

MicroVideo

Models AUC gAUC
50 100 200 500 1000 50 100 200 500 1000
DCNv2 68.89/+0.2% 70.37/+2.4% 70.53/+2.6% 70.75/+2.9% 71.19/+3.6% 70.07/+0.5% 70.42/+1.0% 70.58/+1.2% 70.63/+1.3% 70.73/+1.4%
FinalMLP 68.76/+0.0% 70.05/+1.9% 70.33/+2.3% 70.71/+2.9% 71.05/+3.4% 69.98/+0.3% 70.09/+0.5% 70.27/+0.8% 70.61/+1.2% 70.44/+1.0%
DIN 68.74/+0.0% 70.20/+2.1% 70.57/+2.7% 71.62/+4.2% 71.80/+4.5% 69.74/+0.0% 70.15/+0.6% 70.01/+0.4% 71.18/+2.1% 71.31/+2.3%
DIEN
70.11/+2.0% 70.50/+2.6% 71.19/+3.6% 71.80/+4.5% 72.20/+5.0% 70.63/+1.3% 71.07/+1.9% 71.15/+2.0% 71.32/+2.3% 71.41/+2.4%
TransAct 69.61/+1.3% 70.39/+2.4% 70.62/+2.7% 71.02/+3.3% 71.18/+3.5% 69.64/-0.1% 69.84/+0.1% 69.94/+0.3% 70.22/+0.7% 70.28/+0.8%
SIM 69.91/+1.7% 70.41/+2.4% 70.57/+2.7% 70.76/+2.9% 70.76/+2.9% 69.82/+0.1% 69.82/+0.1% 69.92/+0.3% 70.40/+0.9% 70.38/+0.9%
ETA 69.94/+1.7% 70.45/+2.5% 70.58/+2.7% 70.33/+2.3% 70.31/+2.3% 69.76/+0.0% 69.92/+0.3% 70.02/+0.4% 71.45/+2.5% 71.39/+2.4%
SDIM 69.88/+1.7% 70.81/+3.0% 71.51/+4.0% 71.53/+4.1% 71.73/+4.3% 69.68/-0.1% 69.78/+0.1% 69.84/+0.1% 69.89/+0.2% 69.86/+0.2%
TWIN 69.84/+1.6% 70.54/+2.6% 70.57/+2.7% 71.86/+4.5% 71.89/+4.6% 69.67/-0.1% 69.79/+0.1% 69.91/+0.2% 70.61/+1.2% 70.64/+1.3%
MIRRN 70.05/+1.9% 70.94/+3.2% 71.15/+3.5% 71.31/+3.7% 71.30/+3.7% 69.80/+0.1% 70.05/+0.4% 70.17/+0.6% 70.30/+0.8% 70.30/+0.8%

KuaiVideo

Models AUC gAUC
50 100 200 500 1000 50 100 200 500 1000
DCNv2 68.61/+0.6% 69.16/+1.4% 69.79/+2.3% 69.95/+2.6% 69.99/+2.6% 66.74/+0.0% 67.01/+0.4% 67.17/+0.6% 67.22/+0.7% 67.23/+0.7%
FinalMLP 69.23/+1.5% 69.75/+2.3% 70.31/+3.1% 70.40/+3.2% 70.64/+3.6% 66.90/+0.2% 67.11/+0.5% 67.18/+0.6% 67.28/+0.8% 67.40/+1.0%
DIN 68.21/+0.0% 68.61/+0.6% 68.96/+1.1% 69.55/+2.0% 69.58/+2.0% 66.76/+0.0% 66.89/+0.2% 66.87/+0.2% 66.97/+0.3% 67.12/+0.5%
DIEN 68.67/+0.7% 68.89/+1.0% 69.62/+2.1% 69.73/+2.2% 69.73/+2.2% 66.74/+0.0% 66.79/+0.0% 66.88/+0.2% 66.85/+0.1% 66.84/+0.1%
TransAct 67.90/-0.5% 68.87/+1.0% 68.58/+0.5% 69.13/+1.3% 69.29/+1.6% 66.33/-0.6% 66.49/-0.4% 66.77/+0.0% 66.82/+0.1% 66.81/+0.1%
SIM 68.43/+0.3% 69.04/+1.2% 68.96/+1.1% 68.96/+1.1% 68.95/+1.1% 66.55/-0.3% 66.58/-0.3% 66.75/+0.0% 66.86/+0.1% 66.83/+0.1%
ETA 68.89/+1.0% 69.18/+1.4% 69.50/+1.9% 69.22/+1.5% 69.26/+1.5% 66.66/-0.1% 67.12/+0.5% 67.25/+0.7% 67.38/+0.9% 67.31/+0.8%
SDIM 68.67/+0.7% 69.22/+1.5% 69.58/+2.0% 69.62/+2.1% 69.22/+1.5% 66.72/-0.1% 66.75/+0.0% 66.93/+0.3% 67.10/+0.5% 67.40/+1.0%
TWIN 68.71/+0.7% 69.09/+1.3% 69.09/+1.3% 69.37/+1.7% 69.43/+1.8% 66.65/-0.2% 66.86/+0.1% 67.01/+0.4% 67.28/+0.8% 67.30/+0.8%
MIRRN 68.43/+0.3% 69.12/+1.3% 69.20/+1.5% 69.27/+1.6% 69.29/+1.6% 66.79/+0.0% 66.86/+0.1% 66.93/+0.3% 67.08/+0.5% 67.15/+0.6%

EBNeRD-Small

Models AUC gAUC
50 100 200 500 1000 50 100 200 500 1000
DCNv2 69.86/-0.8% 70.72/+0.4% 70.69/+0.4% 71.44/+1.4% 71.39/+1.3% 69.44/-0.8% 70.31/+0.5% 70.43/+0.6% 70.82/+1.2% 70.82/+1.2%
FinalMLP 69.94/-0.7% 70.75/+0.4% 70.88/+0.6% 71.38/+1.3% 71.51/+1.5% 69.90/-0.1% 70.47/+0.7% 70.50/+0.7% 70.80/+1.2% 70.92/+1.3%
DIN 70.44/+0.0% 71.04/+0.9% 71.20/+1.1% 71.86/+2.0% 71.92/+2.1% 69.98/+0.0% 70.45/+0.7% 70.60/+0.9% 71.23/+1.8% 71.28/+1.9%
DIEN 69.93/-0.7% 70.52/+0.1% 70.88/+0.6% 71.37/+1.3% 70.95/+0.7% 69.80/-0.3% 70.19/+0.3% 70.51/+0.8% 70.91/+1.3% 70.78/+1.1%
TransAct 70.26/-0.3% 70.82/+0.5% 71.14/+1.0% 71.32/+1.2% 71.45/+1.4% 69.87/-0.2% 70.29/+0.4% 70.64/+0.9% 70.71/+1.0% 70.84/+1.2%
SIM 70.33/-0.2% 70.44/+0.0% 70.44/+0.0% 70.04/-0.6% 70.03/-0.6% 69.89/-0.1% 70.02/+0.1% 69.94/-0.1% 69.78/-0.3% 69.54/-0.6%
ETA 70.19/-0.4% 70.50/+0.1% 70.73/+0.4% 70.88/+0.6% 70.82/+0.5% 69.90/-0.1% 70.02/+0.1% 70.07/+0.1% 70.13/+0.2% 70.17/+0.3%
SDIM 70.38/-0.1% 70.42/+0.0% 71.16/+1.0% 71.12/+1.0% 71.41/+1.4% 69.78/-0.3% 69.98/+0.0% 70.55/+0.8% 70.51/+0.8% 70.72/+1.1%
TWIN 70.26/-0.3% 70.26/-0.3% 70.40/-0.1% 70.80/+0.5% 70.50/+0.1% 69.65/-0.5% 69.75/-0.3% 69.86/-0.2% 70.36/+0.5% 69.91/-0.1%
MIRRN 69.46/-1.4% 69.54/-1.3% 69.59/-1.2% 69.81/-0.9% 69.63/-1.1% 69.22/-1.1% 69.29/-1.0% 69.21/-1.1% 69.42/-0.8% 69.69/-0.4%

How to Reproduce

To reproduce the results, please refer to the benchmark folder for the necessary configurations and running logs. For example, you can execute the following script to run SIM on MicroVideo with a sequence length of 50 (where tag=005 indicates the best-tuned hyper-parameters):

python run_param_tuner.py --config benchmark/SIM_microvideo1.7m_x2/SIM_microvideo1.7m_x2_tuner_config_01_50.yaml --tag 005 --gpu 0

About

LongCTR: A Long Sequence Modeling Benchmark for CTR Prediction

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •