AutoCas: Autoregressive Cascade Predictor in Social Networks via Large Language Models

TL;DR

AutoCas addresses cascade popularity prediction under diverse diffusion dynamics and limited data by repurposing decoder-only LLMs as cascade predictors. It tokenizes cascades into sequences, reformulates diffusion as an autoregressive task, and uses cascade prompts to adapt the model while freezing the LLM and training lightweight projection layers. Across Weibo, Twitter, and APS, AutoCas achieves state-of-the-art MSLE and MAPE, with clear gains as LLM size increases, demonstrating scalable performance. The approach enables flexible inference at arbitrary observation times without retraining, offering practical benefits for viral marketing, misinformation control, and recommender systems.

Abstract

Popularity prediction in information cascades plays a crucial role in social computing, with broad applications in viral marketing, misinformation control, and content recommendation. However, information propagation mechanisms, user behavior, and temporal activity patterns exhibit significant diversity, necessitating a foundational model capable of adapting to such variations. At the same time, the amount of available cascade data remains relatively limited compared to the vast datasets used for training large language models (LLMs). Recent studies have demonstrated the feasibility of leveraging LLMs for time-series prediction by exploiting commonalities across different time-series domains. Building on this insight, we introduce the Autoregressive Information Cascade Predictor (AutoCas), an LLM-enhanced model designed specifically for cascade popularity prediction. Unlike natural language sequences, cascade data is characterized by complex local topologies, diffusion contexts, and evolving dynamics, requiring specialized adaptations for effective LLM integration. To address these challenges, we first tokenize cascade data to align it with sequence modeling principles. Next, we reformulate cascade diffusion as an autoregressive modeling task to fully harness the architectural strengths of LLMs. Beyond conventional approaches, we further introduce prompt learning to enhance the synergy between LLMs and cascade prediction. Extensive experiments demonstrate that AutoCas significantly outperforms baseline models in cascade popularity prediction while exhibiting scaling behavior inherited from LLMs. Code is available at this repository: https://anonymous.4open.science/r/AutoCas-85C6

Figures (8)

• Figure 1: Illustration of cascade popularity prediction. Given an information item, the target is to model its diffusion process and predict how many users will aware it.
• Figure 2: The motivation of adapting large language models for cascade modeling, which demonstrates the process of text generation and cascade diffusion, indicating that both can be viewed as sequence modeling from the token perspective.
• Figure 3: The overview framework of AutoCas. (1) Local and global embeddings extraction and cascade token sequences construction; (2) Autoregressive cascade modeling with pre-trained large language model as backbone; (3) Textual prompt template design and injection.
• Figure 4: Ablation study results of AutoCas and its variants on three datasets.
• Figure 5: Performance and training efficiency comparison of AutoCas with different LLMs across three datasets (Weibo: 1 Hour, Twitter: 2 Days, and APS: 5 Years), evaluated under the same batch size and learning rate.