Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PI2I: A Personalized Item-Based Collaborative Filtering Retrieval Framework

Shaoqing Wang, Yingcai Ma, Kairui Fu, Ziyang Wang, Dunxian Huang, Yuliang Yan, Jian Wu

TL;DR

PI2I tackles the limitations of traditional item-based CF and two-tower retrieval by introducing a two-stage framework that first broadens the candidate pool through a relaxed-indexer stage (IBS) and then refines personalization with an interactive, trigger-aware retrieval stage (PRS). The IBS constructs an item-to-item table using Swing-based similarity and a large truncation to maximize Hit Rate, while the PRS employs Trigger-Target sampling and a target-attention–driven model to capture nuanced user-item interactions beyond simple inner products, trained with a negative-likelihood objective. Offline results show PI2I surpasses CF baselines and rivals mainstream Two-Tower models, with strong performance on dense datasets and notable gains at large recall thresholds on sparse data; online deployment on Taobao's homepage yielded a +0.8% improvement in Hit Rate and a +1.05% lift in transactions. A publicly released Taobao dataset of 130 million interactions provides a valuable benchmark for future research, and the work demonstrates practical viability through asynchronous nearline deployment and substantial online gains.

Abstract

Efficiently selecting relevant content from vast candidate pools is a critical challenge in modern recommender systems. Traditional methods, such as item-to-item collaborative filtering (CF) and two-tower models, often fall short in capturing the complex user-item interactions due to uniform truncation strategies and overdue user-item crossing. To address these limitations, we propose Personalized Item-to-Item (PI2I), a novel two-stage retrieval framework that enhances the personalization capabilities of CF. In the first Indexer Building Stage (IBS), we optimize the retrieval pool by relaxing truncation thresholds to maximize Hit Rate, thereby temporarily retaining more items users might be interested in. In the second Personalized Retrieval Stage (PRS), we introduce an interactive scoring model to overcome the limitations of inner product calculations, allowing for richer modeling of intricate user-item interactions. Additionally, we construct negative samples based on the trigger-target (item-to-item) relationship, ensuring consistency between offline training and online inference. Offline experiments on large-scale real-world datasets demonstrate that PI2I outperforms traditional CF methods and rivals Two-Tower models. Deployed in the "Guess You Like" section on Taobao, PI2I achieved a 1.05% increase in online transaction rates. In addition, we have released a large-scale recommendation dataset collected from Taobao, containing 130 million real-world user interactions used in the experiments of this paper. The dataset is publicly available at https://huggingface.co/datasets/PI2I/PI2I, which could serve as a valuable benchmark for the research community.

PI2I: A Personalized Item-Based Collaborative Filtering Retrieval Framework

TL;DR

PI2I tackles the limitations of traditional item-based CF and two-tower retrieval by introducing a two-stage framework that first broadens the candidate pool through a relaxed-indexer stage (IBS) and then refines personalization with an interactive, trigger-aware retrieval stage (PRS). The IBS constructs an item-to-item table using Swing-based similarity and a large truncation to maximize Hit Rate, while the PRS employs Trigger-Target sampling and a target-attention–driven model to capture nuanced user-item interactions beyond simple inner products, trained with a negative-likelihood objective. Offline results show PI2I surpasses CF baselines and rivals mainstream Two-Tower models, with strong performance on dense datasets and notable gains at large recall thresholds on sparse data; online deployment on Taobao's homepage yielded a +0.8% improvement in Hit Rate and a +1.05% lift in transactions. A publicly released Taobao dataset of 130 million interactions provides a valuable benchmark for future research, and the work demonstrates practical viability through asynchronous nearline deployment and substantial online gains.

Abstract

Efficiently selecting relevant content from vast candidate pools is a critical challenge in modern recommender systems. Traditional methods, such as item-to-item collaborative filtering (CF) and two-tower models, often fall short in capturing the complex user-item interactions due to uniform truncation strategies and overdue user-item crossing. To address these limitations, we propose Personalized Item-to-Item (PI2I), a novel two-stage retrieval framework that enhances the personalization capabilities of CF. In the first Indexer Building Stage (IBS), we optimize the retrieval pool by relaxing truncation thresholds to maximize Hit Rate, thereby temporarily retaining more items users might be interested in. In the second Personalized Retrieval Stage (PRS), we introduce an interactive scoring model to overcome the limitations of inner product calculations, allowing for richer modeling of intricate user-item interactions. Additionally, we construct negative samples based on the trigger-target (item-to-item) relationship, ensuring consistency between offline training and online inference. Offline experiments on large-scale real-world datasets demonstrate that PI2I outperforms traditional CF methods and rivals Two-Tower models. Deployed in the "Guess You Like" section on Taobao, PI2I achieved a 1.05% increase in online transaction rates. In addition, we have released a large-scale recommendation dataset collected from Taobao, containing 130 million real-world user interactions used in the experiments of this paper. The dataset is publicly available at https://huggingface.co/datasets/PI2I/PI2I, which could serve as a valuable benchmark for the research community.
Paper Structure (26 sections, 12 equations, 6 figures, 4 tables, 1 algorithm)

This paper contains 26 sections, 12 equations, 6 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Collaborative Filtering
  4. Neural Network Methods
  5. Graph-based Methods
  6. Methodology
  7. Indexer Building Stage
  8. Personalized Retrieval Stage
  9. Trigger-Target Sampling
  10. Light and Efficient Model
  11. Loss Function
  12. Inference
  13. Experiments
  14. Dataset
  15. Experimental Setup
  16. ...and 11 more sections

Figures (6)

  • Figure 1: A general multi-stage architecture in modern recommender systems.
  • Figure 2: (a) Illustration of traditional I2I methods. (b) Illustration of the trigger-target sampling strategy. (c) The overall framework of PI2I . (d) Procedure of the PI2I Online Asynchronous Inference.
  • Figure 3: Ablation study.
  • Figure 4: Truncation Parameter Trade-off.
  • Figure 5: Temporal Decay Pattern of Behavior Triggering.
  • ...and 1 more figures