LISRec: Modeling User Preferences with Learned Item Shortcuts for Sequential Recommendation

TL;DR

LISRec tackles noise in sequential recommendations by extracting stable user preferences from interaction histories through learned semantic shortcuts. It builds per-user semantic graphs using instruction-tuned text representations to compute item similarities, then selects the maximal connected component as a denoised shortcut set to train the recommender. The approach yields up to 13% gains on Yelp and Amazon Product datasets and demonstrates superior denoising performance compared to baselines, with analysis showing improved semantic alignment and robust transfer across architectures. This work highlights the value of explicit, semantically grounded noise filtering for robust, next-item recommendations in real-world settings.

Abstract

User-item interaction histories are pivotal for sequential recommendation systems but often include noise, such as unintended clicks or actions that fail to reflect genuine user preferences. To address this, we propose Learned Item Shortcuts for Sequential Recommendation (LISRec), a novel framework that explicitly captures stable preferences by extracting personalized semantic shortcuts from historical interactions. LISRec first learns task-agnostic semantic representations to assess item similarities, then constructs a personalized semantic graph over all user-interacted items. By identifying the maximal semantic connectivity subset within this graph, LISRec selects the most representative items as semantic shortcuts to guide user preference modeling. This focused representation filters out irrelevant actions while preserving the diversity of genuine interests. Experimental results on the Yelp and Amazon Product datasets illustrate that LISRec achieves a 13% improvement over baseline recommendation models, showing its effectiveness in capturing stable user interests. Further analysis indicates that shortcut-based histories better capture user preferences, making more accurate and relevant recommendations. All codes and datasets are available at https://github.com/NEUIR/LISRec.

Figures (6)

• Figure 1: Illustration of Our LISRec Method. LISRec constructs a personalized semantic interaction graph and extracts shortcuts, representing dominant item subsets that provide denoised user representations for preference modeling.
• Figure 2: The Model Architecture of LISRec.
• Figure 3: Hyperparameter Sensitivity Analysis. We conduct experiments on the threshold hyperparameter, $\tau$, with results derived from the Amazon Beauty dataset using the Bert4Rec \ref{['fig:param:a']} and TASTE \ref{['fig:param:b']} backbone models.
• Figure 4: Performance of Recommendation Models Using Different Semantic Shortcut Extraction Methods.
• Figure 5: Effectiveness of LISRec in Learning Item Representations for Semantic Shortcut Extraction. Figure \ref{['fig:sim:a']} and Figure \ref{['fig:sim:b']} illustrate the item similarities among $M_\text{Filter}$-retained items and the ground truth items.