Large language model tools such as ChatGPT are increasingly used for information seeking in place of traditional search engines. However, limited research is available regarding how lay users search for and evaluate information using ChatGPT. By using scenario-based interviews, this study explores how lay users employ ChatGPT for information seeking in-situ, addressing a significant research gap. Sixteen participants completed a set of search tasks and reflected how they sought and evaluated information. The findings demonstrate that although the information seeking process and techniques resembled those commonly used in other information seeking systems, many techniques were missing. While ChatGPT's responses may reduce cognitive barriers in the information seeking process, lay users face challenges in effectively framing prompts and have concerns over transparency of information source and information quality.

Previous studies have indicated that problematic social media use (PSMU) is positively related to depressive symptoms. However, the potential mechanisms underlying this relationship are not well understood. For example, prior research found that depressive symptoms are related to one’s social network structure, such as network size and density. Therefore, we examined whether one’s social network size and density mediate the relationship between PSMU and depressive symptoms. We conducted an in-person survey to collect measures of PSMU, social network size and density, and depressive symptoms. Our analysis showed that there was a positive relationship between PSMU and depressive symptoms, which was mediated by social network density but not size. Specifically, the greater one’s PSMU, the less dense their social network, and the greater their depressive symptoms. Our research suggests that PSMU may affect how individuals maintain their social connections with others, which may affect mental health. Implications were discussed.

This study evaluates conversational behavior among participants who engaged in conversations with a chatbot or a human support provider. By analyzing 1,232 utterances in the 114 supportive conversations collected from an experiment involving different conditions of chatbots (expert vs. novice bot; high vs. low contingent bot) and a human support provider, the study discovered five conversational sequences, including the extended work-up, the prototypical, the extended delivery, early conclusion, and the close implicature sequence. Comparing human and chatbot conversations, the results showed that conversations with humans were most likely to follow the extended delivery sequence that involved extensive description of the problem, compared to any chatbot conditions. The primary difference between chatbot conditions was that participants were less likely to engage in delivery behavior involving a description of the problem when interacting with the expert chatbot than with the novice chatbot. Theoretical and practical implications were discussed.

Nowadays, the spread of misinformation is a prominent problem in society. Our research focuses on aiding the automatic identification of misinformation by analyzing the persuasive strategies employed in textual documents. We introduce a novel annotation scheme encompassing common persuasive writing tactics to achieve our objective. Additionally, we provide a dataset on health misinformation, thoroughly annotated by experts utilizing our proposed scheme. Our contribution includes proposing a new task of annotating pieces of text with their persuasive writing strategy types. We evaluate fine-tuning and prompt-engineering techniques with pre-trained language models of the BERT family and the generative large language models of the GPT family using persuasive strategies as an additional source of information. We evaluate the effects of employing persuasive strategies as intermediate labels in the context of misinformation detection. Our results show that those strategies enhance accuracy and improve the explainability of misinformation detection models. The persuasive strategies can serve as valuable insights and explanations, enabling other models or even humans to make more informed decisions regarding the trustworthiness of the information.

The emerging uptake of AI chatbots for social support entails systematic comparisons between human and non-human entities as sources of support. In a between-subject experimental study, a human and two types of ostensible chatbots (using a wizard of oz design) had supportive conversations with college students who were experiencing stressful situations during the pandemic. We found that when compared with a less ideal chatbot (i.e., low-contingent chatbot), (1) the human support provider was perceived with more warmth, which directly reduced emotional distress among participants; (2) the ideal chatbot (i.e., high-contingent chatbot) was perceived to be more competent, which activated participants' cognitive reappraisal of their stressful situations and subsequently reduced emotional distress. The human provider and the ideal chatbot did not differ in users' perceived competence or warmth, although the human provider was more effective at activating participants' cognitive reappraisal. This study integrates human communication theories into human-computer interaction work and contributes by positioning and theorizing user perceptions of chatbots in a larger process from support sources with varying communication competence to users' cognitive and emotional responses, and ultimately to the stress outcome. Theoretical and design implications are discussed.