SAMD: An Industrial Framework for Heterogeneous Multi-Scenario Recommendation

Huan, Zhaoxin; Li, Ang; Zhang, Xiaolu; Xu, Min; Yang, Junfang; He, Yong; Zhou, Jun

doi:10.1145/3580305.3599955

Cited by 4 publications

(1 citation statement)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The input to the Domain Adaptive Multi-Task module is the concatenation of representations of user, query and item towers as x = 𝐸 (𝑈 ) ⊕𝐸 (𝑄) ⊕𝐸 (𝐼 ). For multi-domain setting, a series of multi-task and multi-domain models are proposed, such as SharedBottom [3], MMoE [15], PLE [19], STAR [16], SAMD [11], etc. These models use shared structures (Experts or MLP layers) to model the similarity among different tasks or domains, and use individual structures to learn the domain-specific properties.…”

Section: Domain Adaptivementioning

confidence: 99%

An Unified Search and Recommendation Foundation Model for Cold-Start Scenario

Gong,

Ding,

et al. 2023

Proceedings of the 32nd ACM International Conference on Information and Knowledge Management

View full text Add to dashboard Cite

In modern commercial search engines and recommendation systems, data from multiple domains is available to jointly train the multi-domain model. Traditional methods train multi-domain models in the multi-task setting, with shared parameters to learn the similarity of multiple tasks, and task-specific parameters to learn the divergence of features, labels, and sample distributions of individual tasks. With the development of large language models, LLM can extract global domain-invariant text features that serve both search and recommendation tasks. We propose a novel framework called S&R Multi-Domain Foundation, which uses LLM to extract domain invariant features, and Aspect Gating Fusion to merge the ID feature, domain invariant text features and task-specific heterogeneous sparse features to obtain the representations of query and item. Additionally, samples from multiple search and recommendation scenarios are trained jointly with Domain Adaptive Multi-Task module to obtain the multi-domain foundation model. We apply the S&R Multi-Domain foundation model to cold start scenarios in the pretrain-finetune manner, which achieves better performance than other SOTA transfer learning methods. The S&R Multi-Domain Foundation model has been successfully deployed in Alipay Mobile Application's online services, such as content query recommendation and service card recommendation, etc.

show abstract