Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improving Sequential Recommender Systems with Online and In-store User Behavior

Luyi Ma, Aashika Padmanabhan, Anjana Ganesh, Shengwei Tang, Jiao Chen, Xiaohan Li, Lalitesh Morishetti, Kaushiki Nag, Malay Patel, Jason Cho, Sushant Kumar, Kannan Achan

TL;DR

This work tackles the challenge of predicting future online interactions by leveraging in-store user behavior. It introduces a hybrid omnichannel data pipeline to fuse online streaming data with offline in-store transactions and a Store Transaction Encoder that converts in-store item sets into single embeddings compatible with Transformer-based sequential recommender systems. Empirical results on a real-world dataset show that incorporating in-store data improves next-item prediction, with the attention-based encoder delivering the largest gains. The approach enables more accurate, scalable online recommendations in omnichannel retail and supports real-time inference and retraining through a unified feature store.

Abstract

Online e-commerce platforms have been extending in-store shopping, which allows users to keep the canonical online browsing and checkout experience while exploring in-store shopping. However, the growing transition between online and in-store becomes a challenge to sequential recommender systems for future online interaction prediction due to the lack of holistic modeling of hybrid user behaviors (online and in-store). The challenges are twofold. First, combining online and in-store user behavior data into a single data schema and supporting multiple stages in the model life cycle (pre-training, training, inference, etc.) organically needs a new data pipeline design. Second, online recommender systems, which solely rely on online user behavior sequences, must be redesigned to support online and in-store user data as input under the sequential modeling setting. To overcome the first challenge, we propose a hybrid, omnichannel data pipeline to compile online and in-store user behavior data by caching information from diverse data sources. Later, we introduce a model-agnostic encoder module to the sequential recommender system to interpret the user in-store transaction and augment the modeling capacity for better online interaction prediction given the hybrid user behavior.

Improving Sequential Recommender Systems with Online and In-store User Behavior

TL;DR

This work tackles the challenge of predicting future online interactions by leveraging in-store user behavior. It introduces a hybrid omnichannel data pipeline to fuse online streaming data with offline in-store transactions and a Store Transaction Encoder that converts in-store item sets into single embeddings compatible with Transformer-based sequential recommender systems. Empirical results on a real-world dataset show that incorporating in-store data improves next-item prediction, with the attention-based encoder delivering the largest gains. The approach enables more accurate, scalable online recommendations in omnichannel retail and supports real-time inference and retraining through a unified feature store.

Abstract

Online e-commerce platforms have been extending in-store shopping, which allows users to keep the canonical online browsing and checkout experience while exploring in-store shopping. However, the growing transition between online and in-store becomes a challenge to sequential recommender systems for future online interaction prediction due to the lack of holistic modeling of hybrid user behaviors (online and in-store). The challenges are twofold. First, combining online and in-store user behavior data into a single data schema and supporting multiple stages in the model life cycle (pre-training, training, inference, etc.) organically needs a new data pipeline design. Second, online recommender systems, which solely rely on online user behavior sequences, must be redesigned to support online and in-store user data as input under the sequential modeling setting. To overcome the first challenge, we propose a hybrid, omnichannel data pipeline to compile online and in-store user behavior data by caching information from diverse data sources. Later, we introduce a model-agnostic encoder module to the sequential recommender system to interpret the user in-store transaction and augment the modeling capacity for better online interaction prediction given the hybrid user behavior.

Paper Structure

This paper contains 21 sections, 3 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Sequential Recommender Systems
  4. Online and In-store User Behaviors and Item Recommendations
  5. Preliminary
  6. Data Pipeline
  7. Data Sources and Data Ingestion
  8. Data Processing and Transformation
  9. Feature Registry
  10. Online and Offline Stores
  11. Batch and Real-Time Data Use Cases
  12. Models
  13. Store Transaction Encoder with Self-attention
  14. Training and Inference
  15. Evaluation
  16. ...and 6 more sections

Figures (3)

  • Figure 1: A user could have online and in-store shopping behaviors that correlate and reflect the user's shopping interest. Items in online user behavior are differentiable because of individual timestamps, while items from in-store user behaviors usually come in a set with shared timestamps. However, only online user behaviors can't unveil the complete shopping journey.
  • Figure 2: The transformer-based sequential recommender systems cannot process this hybrid sequence because of multiple items at the same timestamp, and we propose an encoder (light-orange box) to enable the modeling of hybrid sequences.
  • Figure 3: Store Transaction Encoder with Self-attention. The attention score is generated from the unnormalized self-attention weight matrix on the given item set, which is permutation invariant. The final encoded item set representation is a weighted sum of the item embedding.