LineRetriever: Planning-Aware Observation Reduction for Web Agents

TL;DR

LineRetriever is a planning-aware retrieval method that selectively reduces web page observations for web agents, maintaining performance while fitting within context limits by focusing on future action relevance.

Contribution

It introduces a novel retrieval approach that explicitly considers planning horizons, improving adaptive web navigation by prioritizing relevant information.

Findings

Reduces observation size without performance loss

Maintains context relevance for future actions

Enhances web agent efficiency in navigation tasks

Abstract

While large language models have demonstrated impressive capabilities in web navigation tasks, the extensive context of web pages, often represented as DOM or Accessibility Tree (AxTree) structures, frequently exceeds model context limits. Current approaches like bottom-up truncation or embedding-based retrieval lose critical information about page state and action history. This is particularly problematic for adaptive planning in web agents, where understanding the current state is essential for determining future actions. We hypothesize that embedding models lack sufficient capacity to capture plan-relevant information, especially when retrieving content that supports future action prediction. This raises a fundamental question: how can retrieval methods be optimized for adaptive planning in web navigation tasks? In response, we introduce \textit{LineRetriever}, a novel approach that leverages a language model to identify and retrieve observation lines most relevant to future navigation steps. Unlike traditional retrieval methods that focus solely on semantic similarity, \textit{LineRetriever} explicitly considers the planning horizon, prioritizing elements that contribute to action prediction. Our experiments demonstrate that \textit{LineRetriever} can reduce the size of the observation at each step for the web agent while maintaining consistent performance within the context limitations.