Persuasion is a central feature of communication, widely used to influence beliefs, attitudes, and behaviors. In today’s digital landscape, across social media and online platforms, persuasive content is pervasive, appearing in political campaigns, marketing, fundraising appeals, and more. These strategies span a broad spectrum, from rational and ethical appeals to highly manipulative tactics, some of which pose significant risks to individuals and society. Despite the growing need to identify and differentiate safe from unsafe persuasion, empirical research in this area remains limited. To address this gap, we introduce SafePersuasion, a two-level taxonomy and annotated dataset that categorizes persuasive techniques based on their safety. We evaluate the baseline performance of three large language models in detecting manipulation and its subtypes, and report only moderate success in distinguishing manipulative content from rational persuasion. By releasing SafePersuasion, we aim to advance research on detecting unsafe persuasion and support the development of tools that promote ethical standards and transparency in persuasive communication online.

This paper describes our approach for the WASSA 2024 Shared Task on Empathy Detection and Emotion Classification and Personality Detection in Interactions at ACL 2024. We focused on Track 3: Empathy Prediction (EMP) which aims to predict the empathy and distress of writers based on their essays. Recently, LLMs have been used to detect the psychological status of the writers based on the texts. Previous studies showed that the performance of LLMs can be improved by designing prompts properly. While diverse approaches have been made, we focus on the fact that LLMs can have different nuances for psychological constructs such as empathy or distress to the specific task. In addition, people can express their empathy or distress differently according to the context. Thus, we tried to enhance the prediction performance of LLMs by proposing a new prompting strategy: Task-Aligned Prompt (TAP). This prompt consists of aligned definitions for empathy and distress to the original paper and the contextual information about the dataset. Our proposed prompt was tested using ChatGPT and GPT4o with zero-shot and few-shot settings and the performance was compared to the plain prompts. The results showed that the TAP-ChatGPT-zero-shot achieved the highest average Pearson correlation of empathy and distress on the EMP track.