*4.2. Automated Facial Expression Response Measurement*

No significant differences in the facial expressions were found. Small sample size could be a reason for the obtained results. Simultaneously, high individual variability in the identified emotions by AFER has previously been reported as an issue in discriminating products [60]. The quasi-absence of the emotions reported through AFER may be attributed to the liquid state of the test samples. Fewer facial movements involved with the liquid state [61,62], absence of apparent emotions [39] evoked by energy drinks or poor emotion recognition by the AFER and higher culture-to-culture (India, China, Cambodia, Vietnam, Korea and Hispanic respondents in this study) or individual-to-individual variances could be, to name a few, some of the reasons for marginal differentiation. Although neutral to positive emotions have been previously found to elicit a few facial expressions [37,63], the efficacy of AFER systems to differentiate between two competing products of a category was not sufficient to replace the existing traditional methods. Pragmatically, products competing in the same category, in general, are not profoundly different from each other, and a high negative valence associated with some of them is not expected. In such cases, the capacity of AFER to differentiate would be limited without the complement of traditional sensory techniques. Culturally specific display rules may also be a reason for the lack of differentiation. These rules govern the amplifying, dampening, or altogether masking of the facial expressions. The use of water, basic taste solutions at varying concentrations and target populations (culturally specific) for calibration or testing may be an option for the efficacy test of AFER systems. A large sample size study using both explicit and implicit methods may shed some light on the topic in the future, but practitioners should complement implicit methods with traditional methods for immediate applications. For future applications, implicit methods can have considerable implications in the retail and foodservice industry, and practitioners should keep testing this methodology.

In comparison to sample A, sample B evoked some negative facial expressions, such as nose scrunch, widen eyes, lip suck, which altogether represent disgust [64] and anger. c Contempt and disgust belong to the family of hostile emotions [65]. Limited facial movement hinders the precise measure of some of the facial expressions such as "fear", "sad" and "anger" [37,63]. Higher attention and engagement observed with sample A may imply that the sample was more enjoyable at the time of tasting. The caffeine, glucose and carbonation in energy drinks enhance the mood and the level of energetic arousal [43] and might be responsible for higher attention and engagement, irrespective of the sample type. More dimpler expressions, though not statistically significant, were observed for sample A. Dimpler has been previously identified as a predictor of positive emotion ratings by machine learning models [66]. In the case of energy drink A, the intensity of dimpler facial movements is higher as compared to energy drink B; thus, exhibiting the higher intensity of emotion contempt during the tasting of energy drink A. The participants elicited a higher intensity of joy and smile in sample B through implicit measures as compared to sample A, which was not the case using the nine-point scale ratings and the explicit study of emotions. Duchenne or genuine smiles are caused by the activation of facial action units 6 and 12 [67], and generally, this is the result of enjoyment and happiness. However, previous studies showed that a smile could be misleading, as many

people smile as a sign of embarrassment [68], disappointment [69] or deliberately to hide emotions [70]. This shows that overall liking, explicit emotional responses, and implicit emotional responses vary in the outcome that they have in relationship to hedonic reactions. Explicit methods of emotion measure the conscious and cognitive actions or associations with the food product [14], whereas implicit methods measure the unconscious responses to the stimuli [71]. This finding was in accordance with the study, which stated that the overall liking, self-reported questionnaire and unconscious responses of the consumers are weakly to moderately correlated [37].
