SynerGen: Contextualized Generative Recommender for Unified Search and Recommendation

2025-09-25

AI Summary

In large-scale recommender systems, the conventional retrieve-then-rank architecture suffers from severe miscalibration and high engineering overhead; meanwhile, existing generative models struggle to jointly support query-based personalized search and query-free recommendation without compromising performance. To address this, we propose SynerGen—the first decoder-only generative recommender model unifying both tasks. Its key innovations include: (1) time-aware rotary positional encoding to enhance temporal modeling of user behavior sequences; and (2) joint optimization of InfoNCE retrieval loss with a hybrid pointwise/pairwise ranking loss, enabling semantic complementarity and end-to-end co-training. Extensive experiments on multiple mainstream benchmarks demonstrate that SynerGen significantly outperforms state-of-the-art generative and joint models, validating the effectiveness and industrial feasibility of a single-lifecycle generative backbone for large-scale information access.

Technology Category

• Data Mining & Knowledge Management: Recommender Systems
• Machine Learning: Deep Generative Models & Autoencoders
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

The dominant retrieve-then-rank pipeline in large-scale recommender systems suffers from mis-calibration and engineering overhead due to its architectural split and differing optimization objectives. While recent generative sequence models have shown promise in unifying retrieval and ranking by auto-regressively generating ranked items, existing solutions typically address either personalized search or query-free recommendation, often exhibiting performance trade-offs when attempting to unify both. We introduce extit{SynerGen}, a novel generative recommender model that bridges this critical gap by providing a single generative backbone for both personalized search and recommendation, while simultaneously excelling at retrieval and ranking tasks. Trained on behavioral sequences, our decoder-only Transformer leverages joint optimization with InfoNCE for retrieval and a hybrid pointwise-pairwise loss for ranking, allowing semantic signals from search to improve recommendation and vice versa. We also propose a novel time-aware rotary positional embedding to effectively incorporate time information into the attention mechanism. extit{SynerGen} achieves significant improvements on widely adopted recommendation and search benchmarks compared to strong generative recommender and joint search and recommendation baselines. This work demonstrates the viability of a single generative foundation model for industrial-scale unified information access.

Problem

• Unifies personalized search and recommendation systems
• Addresses retrieval-ranking performance trade-offs in generative models
• Provides single generative backbone for industrial information access

Innovation

• Single generative backbone for search and recommendation
• Joint optimization with hybrid loss functions
• Time-aware rotary positional embedding in attention