AAAI 2026 Workshop on
New Frontiers in Information Retrieval

This topic focuses on emerging retrieval tasks and benchmarks that go beyond keyword-based or semantic-based document retrieval. We will discuss challenges that involve reasoning-intensive queries, tailored instructions, domain-specific knowledge, and multi-modal content. Moreover, novel retrieval scenarios are emerging, such as cached vector retrieval to accelerate LLM inference, and memory retrieval to support the long-term coherence and decision-making of LLM agent systems.

This topic focuses on novel methods and architectural designs for ranking and reranking. The rapid evolution of foundation models—from BERT to LLMs and diffusion-based language models, and from CLIP to multimodal LLMs and multimodal diffusion models—has opened up significant new opportunities for advancing information retrieval models.

This topic explores recent advancements in data curation and model training strategies. On the data side, we will discuss approaches such as data augmentation to enable new retrieval capabilities (e.g., reasoning and instruction-following), quality assessment and cleaning, and data efficiency for minimal supervision. On the training side, we will discuss techniques such as multi-stage training, smart-batching, negative passage mining, meta-learning, and curriculum learning to improve retrieval performance.

This topic investigates emerging paradigms at the intersection of information retrieval and advanced AI technologies. We will discuss how agent technologies, along with reasoning and planning capabilities, can be leveraged to enhance information retrieval and response generation, including understanding and reformulating user queries, routing requests to appropriate data sources, interleaving reasoning with search tool usage, and processing retrieved content.

In addition to the core themes above, our discussions will extend to critical areas such as robustness against malicious ranking attacks and out-of-distribution data; explainability for responsible and trustworthy systems; and efficiency in the era of neural information retrieval.