Unifying Search and Recommendation with Dual-View Representation Learning in a Generative Paradigm

TL;DR

This work tackles the challenge of unifying search and recommendation by identifying gradient conflicts and manual design complexity as limiting factors in discriminative joint models. It introduces GenSR, a generative framework that uses task-specific prompts to partition the model's parameter space into dedicated subspaces, thereby increasing mutual information between inputs and task outputs. GenSR employs dual-view history representations (CF and semantic) with soft filtering and contrastive alignment, integrated through instruction tuning to generate task-specific outputs for both search and recommendation. Empirical results on two public datasets show GenSR achieves state-of-the-art performance, higher estimated mutual information, and reduced gradient conflicts, validating the proposed information-theoretic rationale and the practical efficacy of the approach.

Abstract

Recommender systems and search engines serve as foundational elements of online platforms, with the former delivering information proactively and the latter enabling users to seek information actively. Unifying both tasks in a shared model is promising since it can enhance user modeling and item understanding. Previous approaches mainly follow a discriminative paradigm, utilizing shared encoders to process input features and task-specific heads to perform each task. However, this paradigm encounters two key challenges: gradient conflict and manual design complexity. From the information theory perspective, these challenges potentially both stem from the same issue -- low mutual information between the input features and task-specific outputs during the optimization process. To tackle these issues, we propose GenSR, a novel generative paradigm for unifying search and recommendation (S&R), which leverages task-specific prompts to partition the model's parameter space into subspaces, thereby enhancing mutual information. To construct effective subspaces for each task, GenSR first prepares informative representations for each subspace and then optimizes both subspaces in one unified model. Specifically, GenSR consists of two main modules: (1) Dual Representation Learning, which independently models collaborative and semantic historical information to derive expressive item representations; and (2) S&R Task Unifying, which utilizes contrastive learning together with instruction tuning to generate task-specific outputs effectively. Extensive experiments on two public datasets show GenSR outperforms state-of-the-art methods across S&R tasks. Our work introduces a new generative paradigm compared with previous discriminative methods and establishes its superiority from the mutual information perspective.

Figures (6)

• Figure 1: (a) and (b) illustrate the discriminative and generative paradigms for unifying S&R, respectively, highlighting their differences in parameter space. While the discriminative paradigm results in low mutual information by optimizing S&R tasks within a shared parameter space, the generative paradigm enhances mutual information by partitioning the space into distinct subspaces through instructions.
• Figure 2: GenSR structure. Given S&R history, GenSR first utilizes dual representation learning to obtain CF and semantic history embeddings. The S&R task unifying module then integrates these dual representations with task-specific prompts into a shared generative model, leveraging contrastive learning for view alignment and instruction tuning to effectively generate task-specific outputs for both search and recommendation.
• Figure 3: Parameter space visualization under KuaiSAR.
• Figure 4: Gradient Conflict Analysis.
• Figure 5: Ablation study of different components in GenSR.

Theorems & Definitions (1)

• Theorem 4.1