Kuaishou Achieves 7 Papers Accepted at AAAI 2025

Kuaishou has announced that 7 of its research papers have been accepted at AAAI 2025, a prestigious international AI conference. The conference, organized by the Association for the Advancement of Artificial Intelligence (AAAI), is considered an A-class academic conference by the China Computer Federation (CCF). AAAI 2025 will be held in Philadelphia, Pennsylvania from February 25 to March 4, 2025, receiving 12,957 submissions with an acceptance rate of 23.4%.

The accepted papers span various cutting-edge AI research areas. Paper 01 introduces D&M, a unified approach for key moment detection and sound effects matching in e-commerce videos, addressing the challenge of adding sound effects to specific moments rather than entire videos. Paper 02 presents a granularity-adaptive spatial evidence tokenization model for video question answering, improving the capture of fine-grained visual content in videos.

Paper 03 proposes LEARN, a large language model-driven knowledge adaptive recommendation framework that combines open-world knowledge with collaborative filtering to enhance recommendation system performance. Paper 04 introduces GCDR, a guided conditional diffusion recommendation model that learns multiple user distributions to better capture user uncertainty and diverse interests.

Paper 05 presents an LLM-powered user simulator for recommender systems, which explicitly simulates user interactions through a two-stage process of understanding product information and evaluating personal interest. Paper 06 introduces KRP (Kuaishou Restoration Processing), specifically focusing on image rescaling using a plug-and-play tri-branch invertible block for video image restoration.

Paper 07 proposes Trigger3, an adaptive model selector for query correction that uses a multi-level approach combining correction triggers, large language model triggers, and fallback triggers to improve search query correction while reducing computational overhead. The research demonstrates Kuaishou's commitment to advancing AI technology and its practical applications in real-world scenarios.