Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TWICE: An LLM Agent Framework for Simulating Personalized User Tweeting Behavior with Long-term Temporal Features

Bingrui Jin, Kunyao Lan, Mengyue Wu

TL;DR

TWICE, an LLM-based framework that leverages the long-term temporal and personalized features of social media data, integrates personalized user profiling, an event-driven memory module, and a workflow for personalized style rewriting, enabling simulation of personalized user tweeting behavior while capturing long-term temporal characteristics.

Abstract

User simulators are often used to generate large amounts of data for various tasks such as generation, training, and evaluation. However, existing approaches concentrate on collective behaviors or interactive systems, struggling with tasks that require modeling temporal characteristics. To address this limitation, we propose TWICE, an LLM-based framework that leverages the long-term temporal and personalized features of social media data. This framework integrates personalized user profiling, an event-driven memory module, and a workflow for personalized style rewriting, enabling simulation of personalized user tweeting behavior while capturing long-term temporal characteristics. In addition, we conduct a comprehensive evaluation with a focus on analyzing tweeting style and event-based changes in behavior. Experiment results demonstrate that our framework improves personalized user simulation by effectively incorporating temporal dynamics, providing a robust solution for long-term behavior tracking.

TWICE: An LLM Agent Framework for Simulating Personalized User Tweeting Behavior with Long-term Temporal Features

TL;DR

TWICE, an LLM-based framework that leverages the long-term temporal and personalized features of social media data, integrates personalized user profiling, an event-driven memory module, and a workflow for personalized style rewriting, enabling simulation of personalized user tweeting behavior while capturing long-term temporal characteristics.

Abstract

User simulators are often used to generate large amounts of data for various tasks such as generation, training, and evaluation. However, existing approaches concentrate on collective behaviors or interactive systems, struggling with tasks that require modeling temporal characteristics. To address this limitation, we propose TWICE, an LLM-based framework that leverages the long-term temporal and personalized features of social media data. This framework integrates personalized user profiling, an event-driven memory module, and a workflow for personalized style rewriting, enabling simulation of personalized user tweeting behavior while capturing long-term temporal characteristics. In addition, we conduct a comprehensive evaluation with a focus on analyzing tweeting style and event-based changes in behavior. Experiment results demonstrate that our framework improves personalized user simulation by effectively incorporating temporal dynamics, providing a robust solution for long-term behavior tracking.
Paper Structure (37 sections, 2 equations, 3 figures, 4 tables)

This paper contains 37 sections, 2 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Tweeting Simulation Agent
  3. Personalized User Profiling
  4. Definition and Scope
  5. General attributes
  6. Personalized attributes
  7. Sampling
  8. Event-driven Memory Module
  9. Memory Structure
  10. Retrieval Strategy
  11. Model-Agnostic Workflow
  12. Event extraction
  13. Stage I: Event-grounded content generation
  14. Stage II: Style adaptation
  15. Evaluation
  16. ...and 22 more sections

Figures (3)

  • Figure 1: LLM Agent Tweeting Simulation Framework TWICE. This framework integrates personalized user profiling, an event-driven memory module, and a workflow for personalized style rewriting.
  • Figure 2: The Procedure of Profile Generation.The profile adopts a two-tier structure that covers both general and personalized attributes
  • Figure 3: Performance Across Four Users' Simulated Tweets.The figure respectively shows the trends of simulation results with respect to four temporal parameters: time_window, state_coeff, memory_num and profile_val.