Generating Plausible and Context-Appropriate Comments on Social Media Posts: A Large Language Model-Based Approach

Research has examined diverse aspects of user commenting behaviors on online platforms, identifying critical engagement factors and influencing marketing strategies. However, a gap remains in predicting the specific comments and replies elicited by individual posts, which has become increasingly relevant as platforms aim to foster positive user interactions. This study applies a large language model (LLM) to bridge this gap, examining the effectiveness of three distinct prompt types in generating realistic comments and replies on social media posts. After fine-tuning LLMs with different prompts and 1,052,288 pairs of posts and comments, LLMs can effectively integrate textual and numerical post features to anticipate user interactions, with significant influence from the nature of the community on the response patterns. These findings underscore the potential of LLMs in enhancing social media analytics and suggest avenues for creating more engaging and responsive online communities. This study discusses the implications of this research for content strategy and highlights critical directions for future inquiry, emphasizing the integration of LLM capabilities with a detailed analysis of the user-generated content.