Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
110 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

A Unified Search and Recommendation Framework Based on Multi-Scenario Learning for Ranking in E-commerce (2405.10835v2)

Published 17 May 2024 in cs.IR

Abstract: Search and recommendation (S&R) are the two most important scenarios in e-commerce. The majority of users typically interact with products in S&R scenarios, indicating the need and potential for joint modeling. Traditional multi-scenario models use shared parameters to learn the similarity of multiple tasks, and task-specific parameters to learn the divergence of individual tasks. This coarse-grained modeling approach does not effectively capture the differences between S&R scenarios. Furthermore, this approach does not sufficiently exploit the information across the global label space. These issues can result in the suboptimal performance of multi-scenario models in handling both S&R scenarios. To address these issues, we propose an effective and universal framework for Unified Search and Recommendation (USR), designed with S&R Views User Interest Extractor Layer (IE) and S&R Views Feature Generator Layer (FG) to separately generate user interests and scenario-agnostic feature representations for S&R. Next, we introduce a Global Label Space Multi-Task Layer (GLMT) that uses global labels as supervised signals of auxiliary tasks and jointly models the main task and auxiliary tasks using conditional probability. Extensive experimental evaluations on real-world industrial datasets show that USR can be applied to various multi-scenario models and significantly improve their performance. Online A/B testing also indicates substantial performance gains across multiple metrics. Currently, USR has been successfully deployed in the 7Fresh App.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (30)
  1. Rich Caruana. 1998. Multitask Learning. In In Learning to Learn. 95–133.
  2. TWIN: TWo-stage Interest Network for Lifelong User Behavior Modeling in CTR Prediction at Kuaishou. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 3785–3794.
  3. Pepnet: Parameter and embedding personalized network for infusing with personalized prior information. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 3795–3804.
  4. Behavior sequence transformer for e-commerce recommendation in alibaba. In Proceedings of the 1st international workshop on deep learning practice for high-dimensional sparse data. 1–4.
  5. A simple framework for contrastive learning of visual representations. In Proceedings of the 37th International Conference on Machine Learning. 1597–1607.
  6. Wide & Deep Learning for Recommender Systems. In Proceedings of the 1st Workshop on Deep Learning for Recommender Systems. 7–10.
  7. DeCLUTR: Deep Contrastive Learning for Unsupervised Textual Representations. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 879–895.
  8. DeepFM: a factorization-machine based neural network for CTR prediction. In Proceedings of the 26th International Joint Conference on Artificial Intelligence. 1725–1731.
  9. Improving multi-scenario learning to rank in e-commerce by exploiting task relationships in the label space. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 2605–2612.
  10. xDeepFM: Combining Explicit and Implicit Feature Interactions for Recommender Systems. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 1754–1763.
  11. Ilya Loshchilov and Frank Hutter. 2019. Decoupled Weight Decay Regularization. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019.
  12. Deep match to rank model for personalized click-through rate prediction. In Proceedings of the AAAI Conference on Artificial Intelligence. 156–163.
  13. Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture-of-Experts. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 1930–1939.
  14. One Model to Serve All: Star Topology Adaptive Recommender for Multi-Domain CTR Prediction. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management. 4104–4113.
  15. When Search Meets Recommendation: Learning Disentangled Search Representation for Recommendation. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1313–1323.
  16. Progressive Layered Extraction (PLE): A Novel Multi-Task Learning (MTL) Model for Personalized Recommendations. In Proceedings of the 14th ACM Conference on Recommender Systems. 269–278.
  17. Laurens van der Maaten and Geoffrey Hinton. 2008. Visualizing Data using t-SNE. In Journal of Machine Learning Research (2008). 2579–2605.
  18. APG: Adaptive Parameter Generation Network for Click-Through Rate Prediction. In Advances in Neural Information Processing Systems. 24740–24752.
  19. AdaSparse: Learning Adaptively Sparse Structures for Multi-Domain Click-Through Rate Prediction. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management. 4635–4639.
  20. USER: A Unified Information Search and Recommendation Model based on Integrated Behavior Sequence. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management. 2373–2382.
  21. Hamed Zamani and W. Bruce Croft. 2018. Joint Modeling and Optimization of Search and Recommendation. In DESIRES’18. 36–41.
  22. Hamed Zamani and W. Bruce Croft. 2020. Learning a Joint Search and Recommendation Model from User-Item Interactions. In Proceedings of the 13th International Conference on Web Search and Data Mining. 717–725.
  23. Leaving No One Behind: A Multi-Scenario Multi-Task Meta Learning Approach for Advertiser Modeling. In Proceedings of the Fifteenth ACM International Conference on Web Search and Data Mining. 1368–1376.
  24. Multiple Relational Attention Network for Multi-task Learning. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 1123–1131.
  25. Joint Learning of E-commerce Search and Recommendation with a Unified Graph Neural Network. In Proceedings of the Fifteenth ACM International Conference on Web Search and Data Mining. 1461–1469.
  26. Deep interest evolution network for click-through rate prediction. In In Proceedings of the AAAI conference on artificial intelligence. 5941–5948.
  27. Deep Interest Network for Click-Through Rate Prediction. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 1059–1068.
  28. Hinet: Novel multi-scenario & multi-task learning with hierarchical information extraction. In 2023 IEEE 39th International Conference on Data Engineering (ICDE). 2969–2975.
  29. S3-Rec: Self-Supervised Learning for Sequential Recommendation with Mutual Information Maximization. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 1893–1902.
  30. Automatic expert selection for multi-scenario and multi-task search. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1535–1544.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (6)
  1. Jinhan Liu (3 papers)
  2. Qiyu Chen (16 papers)
  3. Junjie Xu (23 papers)
  4. Junjie Li (98 papers)
  5. Baoli Li (2 papers)
  6. Sulong Xu (23 papers)

Summary

  • The paper presents the USR framework that integrates search and recommendation by isolating scenario-specific user interests.
  • It leverages dynamic weighting, gating mechanisms, and contrastive learning to fine-tune feature relevance across search and recommendation tasks.
  • Real-world experiments demonstrate improved AUC and conversion metrics, underscoring the framework’s robust impact on e-commerce performance.

A Unified Search and Recommendation Framework for E-commerce

Introduction

Let's dive into a paper proposing a unified approach to Search and Recommendation (S&R) in e-commerce, put forward by a team of researchers at JD.com. Online platforms heavily rely on search and recommendation systems to enhance user experiences by tailoring content to individual preferences. This paper acknowledges the mutual benefits of these two scenarios and presents a novel framework that aims to optimally integrate them.

Traditional vs. Proposed Approach

The conventional multi-scenario learning models have been relying on either shared parameter methodologies or task-specific parameters to address different scenarios independently. While these methods offer a certain level of effectiveness, they fall short in two significant ways:

  1. They miss the finer details of each scenario.
  2. They do not leverage the full potential of the global label space.

To address these issues, the team at JD.com came up with a framework called Unified Search and Recommendation (USR), designed to enhance both search and recommendation functionalities through a more integrated and granular approach.

Key Components of the USR Framework

Here’s a breakdown of the main components of the USR framework:

1. S&R Views User Interest Extractor Layer (IE)

This layer isolates and refines user interests specific to search and recommendation scenarios, rather than generalizing user behaviors across scenarios. By applying dynamic weight techniques and gating mechanisms, the framework can better differentiate between search and recommendation interests. Additionally, contrastive learning is utilized to highlight these differences further.

2. S&R Views Feature Generator Layer (FG)

This layer applies scenario-specific scaling to features that are generally across both S&R scenarios. In simpler terms, it adjusts the importance of each feature based on the specific scenario it is being applied to— either search or recommendation—thereby generating more relevant insights.

3. Global Label Space Multi-Task Layer (GLMT)

Rather than treating the scenarios in isolation, this layer jointly models the primary task (click-through rate or purchase rate) along with auxiliary tasks using conditional probability, ensuring the model benefits from the entire label space across S&R scenarios.

Performance Observations

Strong Numerical Results

The paper reports extensive experimental evaluations on real-world e-commerce datasets, especially from JD.com's 7Fresh App. The results are quite impressive:

  • Significant improvements were noted across multiple Multi-Scenario Learning (MSL) methods:
    • For instance, when the MMoE model was augmented with USR, the AUC (Area Under the Curve) for CTR (Click-Through Rate) improved from 0.8250 to 0.8270.
    • Similar gains were noted in CTCVR (Click-Through Conversion Rate) tasks, demonstrating the framework’s robustness across different metrics.

The improvements can be summarized as follows:

  • A boost in the accuracy of predicting clicks and purchases.
  • Enhanced representation of user interests and scenario-agnostic features.
  • Effective use of global labels to improve auxiliary tasks.

Results and Discussion

The method demonstrated clear benefits when applied to various MSL methods such as SharedBottom, MMoE, PLE, and more, consistently showing improvements in both CTR and CTCVR across both scenarios of search and recommendation. Beyond offline evaluations, online A/B testing in the 7Fresh App showed substantial performance gains in user conversion rates and click-through rates, underscoring the practical utility of the proposed framework.

Implications and Future Perspectives

Practical Implications

  • Enhanced User Experience: By fine-tuning recommendations and search results, users are more likely to find relevant products quickly, increasing overall satisfaction and engagement.
  • Operational Efficiency: With a unified model, e-commerce platforms can reduce redundancy and complexity, potentially leading to cost savings and more efficient use of computational resources.

Theoretical Implications

  • Improved Generalization: The use of scenario-specific layers and global label spaces suggests a pathway toward generalizing multi-task learning models to other paired scenarios beyond just search and recommendation.
  • Contrastive Learning Applications: The innovative use of contrastive learning to differentiate between scenarios provides a rich avenue for further research.

Future Developments

Looking ahead, the incorporation of more advanced techniques, such as reinforcement learning or deeper neural networks, could push the boundaries of what’s possible with unified S&R systems. Also, expanding the framework to handle more complex and varied datasets from different e-commerce contexts could further validate and refine the model's effectiveness.

In summary, the USR framework offers a robust, detailed, and integrated approach to simultaneously enhancing search and recommendation systems on e-commerce platforms. By addressing the fine nuances between these scenarios, it opens up new avenues for both theoretical research and practical applications.

X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets