Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Know Your Intent: An Autonomous Multi-Perspective LLM Agent Framework for DeFi User Transaction Intent Mining

Qian'ang Mao, Yuxuan Zhang, Jiaman Chen, Wenjun Zhou, Jiaqi Yan

TL;DR

The paper tackles understanding user intent behind DeFi transactions in the face of complex on-chain and off-chain signals. It introduces the TIM framework, a structurally deterministic, multi-agent LLM system guided by a grounded-theory–derived DeFi intent taxonomy and a meta-level planner, with reflective multimodal data retrieval and a Cognition Evaluator for verification. Experimental results show TIM substantially outperforms traditional ML models, a single LLM, and single-agent baselines, with ablation indicating each component's necessity. The work enhances explainable, context-aware blockchain analytics and supports applications in risk monitoring, product design, and market forecasting.

Abstract

As Decentralized Finance (DeFi) develops, understanding user intent behind DeFi transactions is crucial yet challenging due to complex smart contract interactions, multifaceted on-/off-chain factors, and opaque hex logs. Existing methods lack deep semantic insight. To address this, we propose the Transaction Intent Mining (TIM) framework. TIM leverages a DeFi intent taxonomy built on grounded theory and a multi-agent Large Language Model (LLM) system to robustly infer user intents. A Meta-Level Planner dynamically coordinates domain experts to decompose multiple perspective-specific intent analyses into solvable subtasks. Question Solvers handle the tasks with multi-modal on/off-chain data. While a Cognitive Evaluator mitigates LLM hallucinations and ensures verifiability. Experiments show that TIM significantly outperforms machine learning models, single LLMs, and single Agent baselines. We also analyze core challenges in intent inference. This work helps provide a more reliable understanding of user motivations in DeFi, offering context-aware explanations for complex blockchain activity.

Know Your Intent: An Autonomous Multi-Perspective LLM Agent Framework for DeFi User Transaction Intent Mining

TL;DR

The paper tackles understanding user intent behind DeFi transactions in the face of complex on-chain and off-chain signals. It introduces the TIM framework, a structurally deterministic, multi-agent LLM system guided by a grounded-theory–derived DeFi intent taxonomy and a meta-level planner, with reflective multimodal data retrieval and a Cognition Evaluator for verification. Experimental results show TIM substantially outperforms traditional ML models, a single LLM, and single-agent baselines, with ablation indicating each component's necessity. The work enhances explainable, context-aware blockchain analytics and supports applications in risk monitoring, product design, and market forecasting.

Abstract

As Decentralized Finance (DeFi) develops, understanding user intent behind DeFi transactions is crucial yet challenging due to complex smart contract interactions, multifaceted on-/off-chain factors, and opaque hex logs. Existing methods lack deep semantic insight. To address this, we propose the Transaction Intent Mining (TIM) framework. TIM leverages a DeFi intent taxonomy built on grounded theory and a multi-agent Large Language Model (LLM) system to robustly infer user intents. A Meta-Level Planner dynamically coordinates domain experts to decompose multiple perspective-specific intent analyses into solvable subtasks. Question Solvers handle the tasks with multi-modal on/off-chain data. While a Cognitive Evaluator mitigates LLM hallucinations and ensures verifiability. Experiments show that TIM significantly outperforms machine learning models, single LLMs, and single Agent baselines. We also analyze core challenges in intent inference. This work helps provide a more reliable understanding of user motivations in DeFi, offering context-aware explanations for complex blockchain activity.

Paper Structure

This paper contains 24 sections, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Blockchain Transaction Analysis
  4. LLM-based Autonomous Agent
  5. Intent Mining Methods and Challenges
  6. TIM: A Structurally Deterministic Workflow Framework for Intent Mining
  7. Problem Formulation
  8. DeFi Transaction Intent Taxonomy Derivation
  9. Reflective Multimodal Live Data Retrieval
  10. Meta-Level Planner (MP)
  11. Perspective-Specific Domain Experts (DE)
  12. Question Solvers (QS)
  13. Cognition Evaluator (CE)
  14. Experiments
  15. Experimental Setup
  16. ...and 9 more sections

Figures (3)

  • Figure 1: User's intent on spot trading for profit. Even though it represents one of the simplest and most common intentions, it still requires analyzing the on-chain data behind three related transactions, summarizing the corresponding semantics, and incorporating external market price data in order to understand the purpose of each transaction and thereby infer the underlying intent.
  • Figure 2: The Transaction Intent Mining (TIM) Framework initiates with a Meta-Level Planner processing transaction parameters and an intent taxonomy to derive perspective-specific domain experts, which then conduct parallel multi-perspective analysis by executing question-solving sequences leveraging live data sources to generate analysis reports, which are aggregated and cognitively evaluated for evidence validity and intent relevance to produce a Final Intent Report detailing validated transaction intents and explanations.
  • Figure 3: The Intent Mining Sequence Diagram illustrates actions (simplified for showcase) within the Case Study.