A Lexical Analysis of online Reviews on Human-AI Interactions

Abstract

The literature review delves into the complex dynamics of human-AI interaction, emphasizing its pivotal role in integrating artificial intelligence (AI) tools successfully. User perceptions of AI, highlighted by Neyazi (2023) and Cinalioglu et al. (2023), significantly influence acceptance and adoption. Trust formation, particularly crucial in domains like healthcare (Davenport & Kalakota, 2019; Wong et al., 2023), underpins widespread AI use. Ethical considerations, including bias and data privacy, demand careful attention (Mittelstadt, 2019; Huriye, 2023). Societal impacts, such as autonomy and collaboration, gain importance (Sankaran et al., 2021; Mirbabaie et al., 2021). In healthcare, addressing concerns is vital for positive outcomes (Esmaeilzadeh et al., 2021; Esmaeilzadeh, 2020), alongside factors like usability (Asan & Choudhury, 2021; Choudhury, 2022). Perceptions of AI risk and trust are influenced by human-likeness (Wong et al., 2023). Understanding user experiences improves outcomes (Rezwana & Maher, 2022; Rezwana, 2022). Despite advancements, understanding user challenges with AI remains incomplete. The review suggests analyzing AI software reviews to identify concerns, aiming to fill this gap and enhance understanding of human-AI interaction's challenges. Through this approach, the research seeks to contribute to the broader discourse on human-AI interaction.The research methodology involves several key steps. Firstly, a comprehensive dataset comprising 55,968 online reviews on AI usage is collected from three prominent websites: G2.com, producthunt.com, and trustpilot.com. These websites are well-established platforms for business software reviews, providing valuable insights into professional AI tool use through user reviews. The collected data is then subjected to rigorous data cleansing procedures using the NLTK library and WordNet Dictionary. This includes removing stop words, extracting adjectives and nouns, combining synonyms and antonyms, and lemmatization, to ensure the quality and consistency of the dataset. A matrix of nouns and adjectives in column and review text created. A weight assigned to each column based on its presence in the reviews (1 if it exists, 0 if not). Subsequently, Exploratory factor analysis is employed to analyze the collected adjectives and nouns and find correlations between words, aiming to identify underlying patterns and group similar traits together. This statistical technique provides valuable insights into the key factors influencing human-AI interaction dynamics, shedding light on the nuanced aspects of this complex relationship.The results obtained from the factor analysis will be further analyzed through content analysis to gain a deeper understanding of the identified factors and their implications for human-AI interaction. Currently, the study is focused on completing the factor analysis, with content analysis planned as the next step in the research process. The findings are expected to provide valuable insights into the critical factors shaping human-AI interaction, contributing to the advancement of our understanding in this domain. This research not only informs future developments in AI technology and user experience but also contributes to ongoing efforts aimed at enhancing human-AI interaction. By adopting the lexical approach and leveraging insights from online reviews, this study aims to bridge the gap between theory and practice in the field of human-AI interaction, paving the way for more user-centric AI systems in the future.