Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Utilization of Cross-scenario Information for Multi-scenario Recommendation

Xiufeng Shu, Ruidong Han, Xiang Li, Wei Lin

TL;DR

This work tackles multi-scenario recommendation under data skew and potential negative transfer by introducing Cross-Scenario Information Interaction (CSII). CSII combines a Transferable Feature Extraction (TFE) module, which learns highly transferable feature representations across scenarios, with a Cross-Scenario Aggregation (CSA) module that uses intra- and inter-scenario attention to adaptively fuse cross-scenario knowledge while mitigating negative transfer. The approach employs scenario-dominated experts plus a shared expert, producing a unified prediction via z = Concat(alpha_s · u_agg + h_s, h_share) and task heads for CTR and CTCVR, with transferability scores and adaptive feature transformations grounded in concrete formulas. Experiments on a large Meituan production dataset and online A/B tests demonstrate superior offline and online performance, including around 1.0% GMV gains overall and up to ~1.5% in extremely sparse scenarios, validating the approach's practical impact.

Abstract

Recommender system of the e-commerce platform usually serves multiple business scenarios. Multi-scenario Recommendation (MSR) is an important topic that improves ranking performance by leveraging information from different scenarios. Recent methods for MSR mostly construct scenario shared or specific modules to model commonalities and differences among scenarios. However, when the amount of data among scenarios is skewed or data in some scenarios is extremely sparse, it is difficult to learn scenario-specific parameters well. Besides, simple sharing of information from other scenarios may result in a negative transfer. In this paper, we propose a unified model named Cross-Scenario Information Interaction (CSII) to serve all scenarios by a mixture of scenario-dominated experts. Specifically, we propose a novel method to select highly transferable features in data instances. Then, we propose an attention-based aggregator module, which can adaptively extract relative knowledge from cross-scenario. Experiments on the production dataset verify the superiority of our method. Online A/B test in Meituan Waimai APP also shows a significant performance gain, leading to an average improvement in GMV (Gross Merchandise Value) of 1.0% for overall scenarios.

Adaptive Utilization of Cross-scenario Information for Multi-scenario Recommendation

TL;DR

This work tackles multi-scenario recommendation under data skew and potential negative transfer by introducing Cross-Scenario Information Interaction (CSII). CSII combines a Transferable Feature Extraction (TFE) module, which learns highly transferable feature representations across scenarios, with a Cross-Scenario Aggregation (CSA) module that uses intra- and inter-scenario attention to adaptively fuse cross-scenario knowledge while mitigating negative transfer. The approach employs scenario-dominated experts plus a shared expert, producing a unified prediction via z = Concat(alpha_s · u_agg + h_s, h_share) and task heads for CTR and CTCVR, with transferability scores and adaptive feature transformations grounded in concrete formulas. Experiments on a large Meituan production dataset and online A/B tests demonstrate superior offline and online performance, including around 1.0% GMV gains overall and up to ~1.5% in extremely sparse scenarios, validating the approach's practical impact.

Abstract

Recommender system of the e-commerce platform usually serves multiple business scenarios. Multi-scenario Recommendation (MSR) is an important topic that improves ranking performance by leveraging information from different scenarios. Recent methods for MSR mostly construct scenario shared or specific modules to model commonalities and differences among scenarios. However, when the amount of data among scenarios is skewed or data in some scenarios is extremely sparse, it is difficult to learn scenario-specific parameters well. Besides, simple sharing of information from other scenarios may result in a negative transfer. In this paper, we propose a unified model named Cross-Scenario Information Interaction (CSII) to serve all scenarios by a mixture of scenario-dominated experts. Specifically, we propose a novel method to select highly transferable features in data instances. Then, we propose an attention-based aggregator module, which can adaptively extract relative knowledge from cross-scenario. Experiments on the production dataset verify the superiority of our method. Online A/B test in Meituan Waimai APP also shows a significant performance gain, leading to an average improvement in GMV (Gross Merchandise Value) of 1.0% for overall scenarios.
Paper Structure (15 sections, 8 equations, 2 figures, 5 tables)

This paper contains 15 sections, 8 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. CROSS-SCENARIO INFORMATION INTERACTION MODEL
  3. Transferable Feature Extraction Module
  4. Cross-Scenario Aggregation Module
  5. Intra-Scenario Transferability Aggregator (Intra-Agg).
  6. Inter-Scenario Transferability Aggregator (Inter-Agg).
  7. Scenario-Dominated Paradigm & Prediction
  8. EXPERIMENTS
  9. Experimental Settings
  10. Overall Experimental Results
  11. Ablation Study
  12. Transferable Feature Extraction Module
  13. Cross-Scenario Aggregation Module
  14. Online A/B Test
  15. CONCLUSION

Figures (2)

  • Figure 1: The structure of Cross-Scenario Information Interaction (CSII), which consists of two key modules: Transferable Feature Extraction (TFE) and Cross-Scenario Aggregation (CSA).
  • Figure 2: The statistics of the score matrix by attention in Inter-agg.