Beyond Faces: AI Facial Emotion Analysis and Focus Groups in Comparative Emotion Evaluation

Introduction

Sentiment testing in qualitative research relies on respondents' perceptions, leading to subjectivity, especially when capturing nuanced emotions. This challenge is particularly pronounced among Asian respondents, who are less expressive. This paper examines the efficacy of AI facial emotion analysis (FEA) using laptop webcams, focusing on accuracy and usability for understanding emotional nuances in reactions to audiovisual materials. By comparing FEA with traditional methods like focus group discussions (FGDs). and interviews, we explore the potential of non-invasive, real-time emotion evaluation, though further validation is needed.

Goals and Methods

The objective was to examine the effectiveness of FEA in identifying emotional nuances compared to traditional sentiment testing through FGDs. FEA was conducted on 6 groups of 8 respondents aged 20 to 79. Respondents were shown 6 videos on laptops, and webcams captured their reactions for FEA via AI analysis through ascribing emotions through movements of facial elements. FGDs followed to gather deeper insights. Outputs from discussions were contrasted with the automated AI outputs of FEA (emotions charted on a line graph).

Results

Conclusions showed using FEA had mixed results. Although non-intrusive and convenient, questions concerning accuracy surfaced when contradictory remarks and false positives from FEA were contrasted with findings from the FGDs. Concerns were also raised over FEA's oversensitivity in identifying emotions, especially in intricate or nuanced situations. Despite these challenges, FEA offers benefits for time-sensitive message testing. Its user-friendly, asynchronous platform allows for larger sample sizes without needing specialized equipment.

Conclusions

This study supports the value of utilizing FEA via laptop webcams to quickly gather organic and unbiased reactions to communication materials compared to running traditional FGDs. While the technology holds promise, limitations in accuracy of capturing emotional responses and understanding reasons are not picked up by the AI tool and still require the supplementary findings from FGDs.