[Paper Review] Generative Recommendation: Towards Next-generation Recommender Paradigm

Wenjie Wang, Xinyu Lin+3 more|arXiv (Cornell University)|Apr 7, 2023

Generative Adversarial Networks and Image SynthesisComputer Science31 citations

TL;DR

The paper proposes GeneRec, a generative recommender paradigm that uses an AI generator to repurpose or create items guided by user instructions and feedback, with fidelity checks and a feasibility study on micro-video content.

ABSTRACT

Recommender systems typically retrieve items from an item corpus for personalized recommendations. However, such a retrieval-based recommender paradigm faces two limitations: 1) the human-generated items in the corpus might fail to satisfy the users' diverse information needs, and 2) users usually adjust the recommendations via inefficient passive feedback, e.g., clicks. Nowadays, AI-Generated Content (AIGC) has revealed significant success, offering the potential to overcome these limitations: 1) generative AI can produce personalized items to satisfy users' information needs, and 2) the newly emerged large language models significantly reduce the efforts of users to precisely express information needs via natural language instructions. In this light, the boom of AIGC points the way towards the next-generation recommender paradigm with two new objectives: 1) generating personalized content through generative AI, and 2) integrating user instructions to guide content generation. To this end, we propose a novel Generative Recommender paradigm named GeneRec, which adopts an AI generator to personalize content generation and leverages user instructions. Specifically, we pre-process users' instructions and traditional feedback via an instructor to output the generation guidance. Given the guidance, we instantiate the AI generator through an AI editor and an AI creator to repurpose existing items and create new items. Eventually, GeneRec can perform content retrieval, repurposing, and creation to satisfy users' information needs. Besides, to ensure the trustworthiness of the generated items, we emphasize various fidelity checks. Moreover, we provide a roadmap to envision future developments of GeneRec and several domain-specific applications of GeneRec with potential research tasks. Lastly, we study the feasibility of implementing AI editor and AI creator on micro-video generation.

Motivation & Objective

• Highlight the potential of AI-generated content (AIGC) to address limitations of retrieval-based recommenders.
• Define the GeneRec paradigm with three components: instructor, AI editor, and AI creator.
• Showcase fidelity checks (bias, privacy, safety, authenticity, legality, identifiability) for generated items.
• Demonstrate feasibility of AI editor/creator on micro-video data and outline future research directions.

Proposed method

• Introduce GeneRec as a two-loop paradigm combining traditional retrieval with an AI generator that uses user instructions and feedback.
• Define three modules: an instructor to process instructions, an AI editor to repurpose existing items, and an AI creator to generate new items.
• Implement fidelity checks covering bias, privacy, safety, authenticity, legality, and identifiability (watermarking, detection).
• Outline evaluation setups: item-side (quality and fidelity) and user-side (satisfaction) assessments.
• Demonstrate tasks for micro-video editing and creation using contemporary AIGC methods (CLIP, RDM, MCVD, VToonify, diffusion models).
• Provide a feasibility study on a large micro-video dataset with experiments on thumbnail selection, thumbnail generation, clipping, content revision, and creation.

Experimental results

Research questions

• RQ1 Can an AI generator effectively repurpose or create items to meet individualized user information needs?
• RQ2 How can user instructions and multimodal feedback guide content generation to improve relevance and satisfaction?
• RQ3 What fidelity and safety mechanisms are feasible to ensure trustworthiness of AI-generated content in recommendations?
• RQ4 Is current AIGC technology capable of supporting practical instantiations of AI editor and AI creator for micro-video content?
• RQ5 What are the potential gains and limitations of integrating generative content into recommender systems compared to traditional retrieval-based methods?

Key findings

• GenRec enables content retrieval, repurposing, and creation to satisfy diverse user needs beyond the existing item corpus.
• Fidelity checks (bias, privacy, safety, authenticity, legality, identifiability) are essential for trustworthy generated content.
• In a micro-video feasibility study, AI editor tasks (thumbnail selection/generation, clipping, content revision) show performance gains over baselines, with RDM/Large diffusion-based methods performing best for some tasks.
• AI editor using CLIP for personalized thumbnail selection improves Cosine@K and PS@K metrics over Random/Original baselines; RDM yields best results for thumbnail generation.
• Micro-video clipping via CLIP consistently outperforms baselines in Cosine@K and PS@K metrics.
• MCVD-based micro-video content revision and creation can improve user-aligned results when guided by user embeddings or historical preferences, though some quality gaps remain compared to unconditional generation.
• Creation tasks show higher short-term cosine similarity but larger FVD scores, indicating current limitations in fully satisfying personalized generation.