SuperCone: Modeling Heterogeneous Experts with Concept Meta-learning for Unified Predictive Segments System

03/09/2022
by   Keqian Li, et al.
0

Understanding users through predicative segments play an essential role for modern enterprises for more efficient and efficient information exchange. For example, by predicting whether a user has particular interest in a particular area of sports or entertainment, we can better serve the user with more relevant and tailored content. However, there exists a large number of long tail prediction tasks that are hard to capture by off the shelf model architectures due to data scarcity and task heterogeneity. In this work, we present SuperCone, our unified predicative segments system that addresses the above challenges. It builds on top of a flat concept representation that summarizes each user's heterogeneous digital footprints, and uniformly models each of the prediction task using an approach called "super learning ", that is, combining prediction models with diverse architectures or learning method that are not compatible with each other or even completely unknown. Following this, we provide end to end deep learning architecture design that flexibly learns to attend to best suited heterogeneous experts while at the same time learns deep representations of the input concepts that augments the above experts by capturing unique signal. Experiments show that SuperCone can outperform state-of-the-art recommendation and ranking algorithms on a wide range of predicative segment tasks, as well as several public structured data learning benchmarks.

READ FULL TEXT

page 8

page 9

page 13

research
03/09/2022

MetaCon: Unified Predictive Segments System with Trillion Concept Meta-Learning

Accurate understanding of users in terms of predicative segments play an...
research
02/22/2018

Federated Meta-Learning for Recommendation

Recommender systems have been widely studied from the machine learning p...
research
06/18/2022

EEML: Ensemble Embedded Meta-learning

To accelerate learning process with few samples, meta-learning resorts t...
research
03/08/2022

Reinforced MOOCs Concept Recommendation in Heterogeneous Information Networks

Massive open online courses (MOOCs), which provide a large-scale interac...
research
08/01/2020

Relation-aware Meta-learning for Market Segment Demand Prediction with Limited Records

Recently, E-commerce platforms have extensive impacts on our human life....
research
08/07/2020

Neural Complexity Measures

While various complexity measures for diverse model classes have been pr...

Please sign up or login with your details

Forgot password? Click here to reset