A Two-Stage Prompt Learning Method for Jointly Predicting Topic and Personality

Yilin Wu, Minjie Yuan, Yuxuan Song, Liping Chen, Chenyu Yuan, Qiudan Li

Abstract: Accurate personality prediction can help management departments analyze users' behaviors and make informed decisions effectively. Existing text-based personality prediction studies mainly rely on deep neural networks or pre-trained language models to extract semantic information and personality traits. However, the text's topic and label description may provide additional personality clues. This paper proposes a topic and personality prediction method based on a large language model (LLM), which utilizes a two-stage prompt strategy to mine the interaction between the topic and personality information. Additionally, labels' descriptions are incorporated to construct cue-based prompts, and a fine-tuning approach is adopted to optimize the model's performance. Experiments on two datasets show the efficacy of the proposed model.

External IDs:dblp:conf/isi/WuYSCYL23