LLM-Explorer: Towards Efficient and Affordable LLM-based Exploration for Mobile Apps

2025-05-15

AI Summary

Existing LLM-driven mobile application exploration methods invoke large language models (LLMs) at every step to generate actions, incurring prohibitive token consumption and computational overhead; moreover, most UI interactions do not require LLM capabilities and are susceptible to LLM biases. Method: We propose a knowledge-centric lightweight exploration paradigm that repositions the LLM as a dynamic knowledge maintainer—not an action generator—enabling LLM-free action selection. Our approach leverages state abstraction, compact knowledge representation, rule- and heuristic-based action generation, and online knowledge refinement. Contribution/Results: Evaluated on 20 representative mobile apps, our method achieves the highest coverage and fastest exploration speed among baselines. It reduces token cost by 148× compared to state-of-the-art approaches, while demonstrating superior efficiency, robustness, and scalability.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Large language models (LLMs) have opened new opportunities for automated mobile app exploration, an important and challenging problem that used to suffer from the difficulty of generating meaningful UI interactions. However, existing LLM-based exploration approaches rely heavily on LLMs to generate actions in almost every step, leading to a huge cost of token fees and computational resources. We argue that such extensive usage of LLMs is neither necessary nor effective, since many actions during exploration do not require, or may even be biased by the abilities of LLMs. Further, based on the insight that a precise and compact knowledge plays the central role for effective exploration, we introduce LLM-Explorer, a new exploration agent designed for efficiency and affordability. LLM-Explorer uses LLMs primarily for maintaining the knowledge instead of generating actions, and knowledge is used to guide action generation in a LLM-less manner. Based on a comparison with 5 strong baselines on 20 typical apps, LLM-Explorer was able to achieve the fastest and highest coverage among all automated app explorers, with over 148x lower cost than the state-of-the-art LLM-based approach.

Problem

• Reducing high token fees in LLM-based mobile app exploration
• Minimizing unnecessary LLM usage for UI action generation
• Achieving efficient app coverage with lower computational cost

Innovation

• Uses LLMs for maintaining knowledge, not actions
• Guides action generation without LLMs
• Achieves high coverage with lower cost