DynaTree: Dynamic Agentic Retrieval Tree for Time-Sensitive News Retrieval

2026-05-29

AI Summary

This work addresses the high inference cost and poor adaptability of existing retrieval-augmented generation (RAG) methods in time-sensitive news retrieval, which stem from the tight coupling of semantic expansion and retrieval decisions. To overcome these limitations, we propose DynaTree, a two-stage framework that decouples reasoning from retrieval: it first constructs an offline, reusable, structure-aware semantic retrieval tree, and then performs lightweight daily selection of temporally localized subtrees for online inference—enabling efficient, training-free dynamic adaptation. This approach effectively balances coverage, freshness, and relevance. Evaluated on the Syft news benchmark and BEIR datasets, DynaTree significantly outperforms standard RAG and state-of-the-art intelligent retrieval methods; online A/B tests further demonstrate improved survival rates of 0.59–0.73, consistently surpassing the current production recall system on a daily basis.

Abstract

Agentic Retrieval-Augmented Generation improves retrieval by integrating planning, tool use, and iterative reasoning, but existing agentic RAG methods often couple semantic expansion with retrieval decisions in short-horizon inference loops, leading to high inference cost and limited suitability for time-sensitive news retrieval. We propose DynaTree, a two-stage framework for efficient and adaptive news retrieval. In the offline stage, DynaTree uses coordinated agents to construct a reusable retrieval tree that materializes the semantic space of a query topic. In the online stage, DynaTree performs lightweight daily subtree selection over a time-localized evaluation proxy, without further agentic reasoning, tree modification, or retraining. Experiments on a multi-day Syft news benchmark and multiple BEIR datasets show that DynaTree achieves strong recall and ranking performance, consistently outperforming standard RAG and prior agentic baselines. We further deploy DynaTree in the Syft production system and evaluate it through online A/B testing from Jan. 28 to Feb. 6, 2026. The dynamically adapted variant improves survival rate from 0.32-0.53 to 0.59-0.73 over a fixed offline-selected subtree and outperforms existing production recallers on every evaluation day. These results show that persistent, structure-aware semantic expansion can translate offline agentic reasoning into practical improvements in coverage, freshness, and relevance for real-world news retrieval.

Problem

• Agentic Retrieval-Augmented Generation
• time-sensitive news retrieval
• semantic expansion
• retrieval efficiency
• inference cost

Innovation

• Agentic Retrieval-Augmented Generation
• Dynamic Retrieval Tree
• Time-Sensitive News Retrieval
• Two-Stage Framework
• Semantic Expansion