.png){kind=icon}
.png){kind=icon}
Return
FLSEST: CTR model based on important features and soft threshold
.png){kind=icon}
OA
AI
DOI：10.15837/ijccc.2026.1.7069.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Click-through rate (CTR) prediction plays a pivotal role in developing effective recommendation systems across industries. While existing models like DeepFM primarily focus on low-order and high-order feature interactions, they often fail to sufficiently account for the heterogeneous importance distribution among individual features. To address this limitation, we propose FLSEST, a novel architecture integrating a squeeze-excitation and soft threshold (SEST) mechanism that dynamically amplifies discriminative features while suppressing noise from less informative ones. Drawing inspiration from FLEN's design philosophy, we additionally introduce a feature-weighting bilinear interaction (FWBI) layer to resolve gradient coupling phenomena during feature interaction learning. Extensive experimental evaluations on multiple public datasets demonstrate that our FLSEST model achieves superior prediction performance compared to state-of-the-art shallow and deep recommendation models. Moreover, integrating our proposed SEST network with mainstream models such as FwFM and DeepFM further enhances their predictive capabilities, confirming the versatility and effectiveness of our approach.
Keywords:
Recommendation algorithm
Click-through rate
Deep learning
Attention mecha-nism
Squeeze excitation and soft threshold network
Journal
I
IF:
1.9
Papers:
28
Citations:
0