Coffee Label Assessment Using Sensory and Biometric Analysis of Self-Isolating Panelists through Videoconference

: Label concepts, information, logos, ﬁgures, and colors of beverages are critical for consumer perception, preference, and purchase intention. This is especially relevant for new beverage products. During social isolation, many sensory laboratories were unable to provide services, making virtual sensory sessions relevant to studying different label concepts and design preferences among consumers. This study proposed a novel virtual sensory system to analyze coffee labels using videoconference, self-reported, and biometric analysis software from video recordings to obtain sensory and emotional responses from 69 participants (power analysis: 1 − β > 0.99) using six different label concepts: (i) fun, (ii) bold, (iii) natural, (iv) everyday, (v) classic, and (vi) premium. The results show that the label concept rated as having the highest perceived quality was premium, presenting signiﬁcant differences ( p < 0.05) compared to all of the other concepts. The least perceived quality score was attributed to the bold concept due to the confronting aroma lexicon (cheese dip), which is supported by previous studies. Furthermore, even though graphics, colors, and the product name can be considered positive attributes, they do not determine perceived quality or purchase intention, which was found for the bold, everyday, and classic concepts. The ﬁndings from this study were as expected and are consistent with those from similar publications related to labels, which shows that the proposed virtual method for sensory sessions and biometrics is reliable. Further technology has been proposed to use this system with multiple participants, which could help beverage companies perform virtual sensory analysis of new products’ labels.


Introduction
Packaging and the information presented on labels are the first points of contact between products and consumers, especially for new products and those unfamiliar to consumers. Therefore, packaging and labels play a significant role in determining preference, liking, and purchase intention [1,2]. Sensory analysis of packaging, labels, and label information is the easiest assessment that can be conducted in social isolation and when products are not as physically available for other sensory assessments. Packaging, labels, and label information can be presented to panelists and consumers through images as stimuli, which have been shown to render similar results statistically compared to those when the packaging is physically available [3] and compared to the sensory characteristics of other products, such as chocolate [4], yogurt [5], beef [6], and baby formula [3]. Furthermore, label designs are one of the most important factors defining the preference and purchase intention of consumers toward wines in general [7], consumers with low involvement [8], and millennials [9], especially when assessing labels with recognizable features, such as animals, novelty designs [7], and colors [10].

Consumer Sensory Session
A virtual sensory session was conducted with 69 participants (Age: 21-53 years old; 78% female, 22% male) recruited from the staff and students from The University of Mel-

Consumer Sensory Session
A virtual sensory session was conducted with 69 participants (Age: 21-53 years old; 78% female, 22% male) recruited from the staff and students from The University of Melbourne (UoM), Australia, and e-mails to personal contacts; participants were regular coffee consumers (at least once a week). According to the Power analysis conducted using SAS ® v. 9.4 (SAS Institute, Cary, NC, USA), the number of participants was statistically sufficient to find significant differences between samples (1 − β > 0.99). The sessions lasted 10-15 min per participant and were conducted via Zoom (Zoom Video Communications, Inc., San Jose, CA, USA) using Google Forms (Google, LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor the session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before the session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were also instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant before the session explaining these instructions with subtitles for non-native English speakers. The testing setting of each participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for all participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,22], FaceScale (emotional response), perceived quality, and willingness to purchase (Table 1), as well as check all that apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2). bourne (UoM), Australia, and e-mails to personal contacts; participants were regular coffee consumers (at least once a week). According to the Power analysis conducted using SAS ® v. 9.4 (SAS Institute, Cary, NC, USA), the number of participants was statistically sufficient to find significant differences between samples (1 − β > 0.99). The sessions lasted 10-15 min per participant and were conducted via Zoom (Zoom Video Communications, Inc., San Jose, CA, USA) using Google Forms (Google, LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor the session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before the session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were also instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant before the session explaining these instructions with subtitles for non-native English speakers. The testing setting of each participant was verified before the researcher started the sessions to initiate the test. The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for all participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,22], FaceScale (emotional response), perceived quality, and willingness to purchase (Table 1), as well as check all that apply (CATA) questions for emojis and preferred areas of interest from label elements (Table 2).

Video Analysis to Obtain Biometrics
All videos from participants recorded per sample were analyzed in batch using an automatic computer application developed by the Digital Agriculture Food and Wine Perceived quality PQuality 9-point scale 1: Extremely low-9: Extremely high Willingness to purchase WPurchase 9-point scale 1: Extremely unlikely-9: Extremely likely Table 2. Options provided for the check all that apply (CATA) questions for assessment of the coffee pod label samples.

CATA Emojis
were conducted via Zoom (Zoom Video Communications, Inc., San Jose, CA, USA) using Google Forms (Goo LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor t session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before th session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant befor the session explaining these instructions with subtitles for non-native English speakers. The testing setting of ea participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for all participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,2 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check all th apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).  Table 2. Options provided for the check all that apply (CATA) questions for assessment of the coffee pod labe samples.

CATA Emojis
were conducted via Zoom (Zoom Video Communications, Inc., San Jose, CA, USA) using Google Forms (Goo LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor t session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before t session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant befo the session explaining these instructions with subtitles for non-native English speakers. The testing setting of ea participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for al participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,2 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check all t apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
were conducted via Zoom (Zoom Video Communications, Inc., San Jose, CA, USA) using Google Forms (G LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monito session in the background and record videos of the participants while looking at the label (15 s) to obtain thei biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Befor session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They we instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant be the session explaining these instructions with subtitles for non-native English speakers. The testing setting of participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check al apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).  Table 2. Options provided for the check all that apply (CATA) questions for assessment of the coffee pod la samples.

CATA Emojis
were conducted via Zoom (Zoom Video Communications, Inc., San Jose, CA, USA) using Google Forms (Go LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They wer instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant bef the session explaining these instructions with subtitles for non-native English speakers. The testing setting of participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for a participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check al apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor t session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before th session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant befor the session explaining these instructions with subtitles for non-native English speakers. The testing setting of ea participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for all participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,2 FaceScale (emotional response), perceived quality, and willingness to purchase (Table 1), as well as check all th apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor t session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before t session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant befo the session explaining these instructions with subtitles for non-native English speakers. The testing setting of ea participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for al participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,2 FaceScale (emotional response), perceived quality, and willingness to purchase (Table 1), as well as check all t apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monito session in the background and record videos of the participants while looking at the label (15 s) to obtain thei biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Befor session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They we instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant be the session explaining these instructions with subtitles for non-native English speakers. The testing setting of participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21 FaceScale (emotional response), perceived quality, and willingness to purchase (Table 1), as well as check a apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).  Table 2. Options provided for the check all that apply (CATA) questions for assessment of the coffee pod la samples.

CATA Emojis
LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monito session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They wer instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant bef the session explaining these instructions with subtitles for non-native English speakers. The testing setting of participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for a participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check al apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor t session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before th session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant befor the session explaining these instructions with subtitles for non-native English speakers. The testing setting of ea participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for all participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,2 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check all th apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor t session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before t session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant befo the session explaining these instructions with subtitles for non-native English speakers. The testing setting of ea participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for al participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,2 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check all t apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monito session in the background and record videos of the participants while looking at the label (15 s) to obtain thei biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Befor session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They we instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant be the session explaining these instructions with subtitles for non-native English speakers. The testing setting of participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check al apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).  Table 2. Options provided for the check all that apply (CATA) questions for assessment of the coffee pod la samples.

CATA Emojis
LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monito session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They wer instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant bef the session explaining these instructions with subtitles for non-native English speakers. The testing setting of participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for a participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check al apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before th session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant befor the session explaining these instructions with subtitles for non-native English speakers. The testing setting of ea participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for all participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,2 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check all th apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before t session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant befo the session explaining these instructions with subtitles for non-native English speakers. The testing setting of ea participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for al participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,2 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check all t apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
session in the background and record videos of the participants while looking at the label (15 s) to obtain thei biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Befor session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They we instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant be the session explaining these instructions with subtitles for non-native English speakers. The testing setting of participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check a apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).  Table 2. Options provided for the check all that apply (CATA) questions for assessment of the coffee pod la samples.

CATA Emojis
session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They wer instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant bef the session explaining these instructions with subtitles for non-native English speakers. The testing setting of participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for a participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21 FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check al apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

Video Analysis to Obtain Biometrics
All videos from participants recorded per sample were analyzed in batch using an automatic computer applicatio developed by the Digital Agriculture Food and Wine (DAFW) group based on the Affectiva (Affectiva, Boston, MA, USA) software development kit (SDK). This application is able to assess participants' facial expressions fr videos using a histogram of the oriented gradient algorithms, which are automatically translated into emotions a related emojis (

Video Analysis to Obtain Biometrics
All videos from participants recorded per sample were analyzed in batch using an automatic computer applicati developed by the Digital Agriculture Food and Wine (DAFW) group based on the Affectiva (Affectiva, Boston MA, USA) software development kit (SDK). This application is able to assess participants' facial expressions f videos using a histogram of the oriented gradient algorithms, which are automatically translated into emotions a related emojis (

Video Analysis to Obtain Biometrics
All videos from participants recorded per sample were analyzed in batch using an automatic computer application developed by the Digital Agriculture Food and Wine (DAFW) group based on the Affectiva (Affectiva, Boston, MA, USA) software development kit (SDK). This application is able to assess participants' facial expressions from videos using a histogram of the oriented gradient algorithms, which are automatically translated into  (Table 3) using support vector machine modeling algorithms in a batch analysis fashion [19,23]. All videos from participants recorded per sample were analyzed in batch using an automatic computer applicat developed by the Digital Agriculture Food and Wine (DAFW) group based on the Affectiva (Affectiva, Boston MA, USA) software development kit (SDK). This application is able to assess participants' facial expressions videos using a histogram of the oriented gradient algorithms, which are automatically translated into emotions related emojis (

Statistical Analysis
All quantitative data (Tables 1 and 3) were analyzed through ANOVA to assess significant differences among samples (p < 0.05), with the least significant differences post hoc test (α = 0.05) using XLSTAT ver. 2020.3.1 (Addinsoft, New York, NY, USA). Frequency data were analyzed using Cochran's Q test (p < 0.05) and McNe with Bonferroni correction post hoc test for pairwise comparison using XLSTAT.
The XLSTAT software was also used to conduct multivariate data analysis as multiple factor analysis (MFA) w mixed data (quantitative and frequencies) based on correlations between variables and factors using all parame assessed in the sensory session (Tables 1-3) and eliminating those whose correlation coefficient (r) did not Emoji (Smiley)

Surprise Emotion
All videos from participants recorded per sample were analyzed in batch using an automatic computer applicat developed by the Digital Agriculture Food and Wine (DAFW) group based on the Affectiva (Affectiva, Boston MA, USA) software development kit (SDK). This application is able to assess participants' facial expressions videos using a histogram of the oriented gradient algorithms, which are automatically translated into emotions related emojis (

Statistical Analysis
All quantitative data (Tables 1 and 3) were analyzed through ANOVA to assess significant differences among samples (p < 0.05), with the least significant differences post hoc test (α = 0.05) using XLSTAT ver. 2020.3.1 (Addinsoft, New York, NY, USA). Frequency data were analyzed using Cochran's Q test (p < 0.05) and McNe with Bonferroni correction post hoc test for pairwise comparison using XLSTAT.
The XLSTAT software was also used to conduct multivariate data analysis as multiple factor analysis (MFA) w mixed data (quantitative and frequencies) based on correlations between variables and factors using all parame assessed in the sensory session (Tables 1-3) and eliminating those whose correlation coefficient (r) did not Fear Emotion developed by the Digital Agriculture Food and Wine (DAFW) group based on the Affectiva (Affectiva, Boston MA, USA) software development kit (SDK). This application is able to assess participants' facial expressions videos using a histogram of the oriented gradient algorithms, which are automatically translated into emotions related emojis (

Statistical Analysis
All quantitative data (Tables 1 and 3) were analyzed through ANOVA to assess significant differences among samples (p < 0.05), with the least significant differences post hoc test (α = 0.05) using XLSTAT ver. 2020.3.1 (Addinsoft, New York, NY, USA). Frequency data were analyzed using Cochran's Q test (p < 0.05) and McNe with Bonferroni correction post hoc test for pairwise comparison using XLSTAT.
The XLSTAT software was also used to conduct multivariate data analysis as multiple factor analysis (MFA) w mixed data (quantitative and frequencies) based on correlations between variables and factors using all parame assessed in the sensory session (Tables 1-3) and eliminating those whose correlation coefficient (r) did not Disgust Emotion MA, USA) software development kit (SDK). This application is able to assess participants' facial expressions videos using a histogram of the oriented gradient algorithms, which are automatically translated into emotions related emojis (

Statistical Analysis
All quantitative data (Tables 1 and 3) were analyzed through ANOVA to assess significant differences among samples (p < 0.05), with the least significant differences post hoc test (α = 0.05) using XLSTAT ver. 2020.3.1 (Addinsoft, New York, NY, USA). Frequency data were analyzed using Cochran's Q test (p < 0.05) and McNe with Bonferroni correction post hoc test for pairwise comparison using XLSTAT.
The XLSTAT software was also used to conduct multivariate data analysis as multiple factor analysis (MFA) w mixed data (quantitative and frequencies) based on correlations between variables and factors using all parame assessed in the sensory session (Tables 1-3) and eliminating those whose correlation coefficient (r) did not Joy Emotion MA, USA) software development kit (SDK). This application is able to assess participants' facial expressions videos using a histogram of the oriented gradient algorithms, which are automatically translated into emotions related emojis (

Statistical Analysis
All quantitative data (Tables 1 and 3) were analyzed through ANOVA to assess significant differences among samples (p < 0.05), with the least significant differences post hoc test (α = 0.05) using XLSTAT ver. 2020.3.1 (Addinsoft, New York, NY, USA). Frequency data were analyzed using Cochran's Q test (p < 0.05) and McNe with Bonferroni correction post hoc test for pairwise comparison using XLSTAT.
The XLSTAT software was also used to conduct multivariate data analysis as multiple factor analysis (MFA) w mixed data (quantitative and frequencies) based on correlations between variables and factors using all parame assessed in the sensory session (Tables 1-3) and eliminating those whose correlation coefficient (r) did not videos using a histogram of the oriented gradient algorithms, which are automatically translated into emotions related emojis (

Statistical Analysis
All quantitative data (Tables 1 and 3) were analyzed through ANOVA to assess significant differences among samples (p < 0.05), with the least significant differences post hoc test (α = 0.05) using XLSTAT ver. 2020.3.1 (Addinsoft, New York, NY, USA). Frequency data were analyzed using Cochran's Q test (p < 0.05) and McNe with Bonferroni correction post hoc test for pairwise comparison using XLSTAT.
The XLSTAT software was also used to conduct multivariate data analysis as multiple factor analysis (MFA) w mixed data (quantitative and frequencies) based on correlations between variables and factors using all parame assessed in the sensory session (Tables 1-3) and eliminating those whose correlation coefficient (r) did not

Statistical Analysis
All quantitative data (Tables 1 and 3) were analyzed through ANOVA to assess significant differences among samples (p < 0.05), with the least significant differences post hoc test (α = 0.05) using XLSTAT ver. 2020.3.1 (Addinsoft, New York, NY, USA). Frequency data were analyzed using Cochran's Q test (p < 0.05) and McNemar with Bonferroni correction post hoc test for pairwise comparison using XLSTAT.
The XLSTAT software was also used to conduct multivariate data analysis as multiple factor analysis (MFA) with mixed data (quantitative and frequencies) based on correlations between variables and factors using all parameters assessed in the sensory session (Tables 1-3) and eliminating those whose correlation coefficient (r) did not contribute much to both factors (r < 0.35 in both F1 and F2). Furthermore, a customized Matlab ® R2020b (Mathworks Inc., Natick, MA, USA) code was used to develop a matrix using the quantitative data (Tables 1 and 3) to assess only the significant correlations (p < 0.05) between the self-reported and biometric responses.

Biometrics for Emotional Responses
Non-significant differences (p > 0.05) were found between the six coffee labels for any of the biometric emotional responses from participants; the means and standard error are shown in Supplementary Material Table S1. On the other hand, Table 4 shows that all attributes from the acceptance test self-reported responses had significant differences (p < 0.05) between samples. The bold label was rated as the lowest in perceived strength (4.27), pleasantness (4.20), FaceScale (4.19), perceived quality (PQuality) (4.55), and willingness to purchase (WPurchase) (3.70), while the natural label was rated as the lowest in arousal (3.75). On the other hand, the premium label was rated the highest in strength (8.48), arousal (6.27), and PQuality (7.95). Furthermore, premium was non-significantly different to natural in pleasantness (6.77 and 7.31, respectively), FaceScale (6.94 and 7.09, respectively), and WPurchase (6.61 and 6.80, respectively). Table 5 shows that there were non-significant differences (p > 0.05) between samples in the selection of emojis, such as . Before the session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were also instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant before the session explaining these instructions with subtitles for non-native English speakers. The testing setting of each participant was verified before the researcher started the sessions to initiate the test.

Emotional Responses Based on Emojis
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for all participants; the order presented is shown in Figure 1.

CATA Emojis
, apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

Video Analysis to Obtain Biometrics
All videos from participants recorded per sample were analyzed in batch using developed by the Digital Agriculture Food and Wine (DAFW) group based on MA, USA) software development kit (SDK). This application is able to assess videos using a histogram of the oriented gradient algorithms, which are autom related emojis (Table 3) using support vector machine modeling algorithms in biometrics through facial recognition to further assess their emotional subcons session, the participants were asked to read the plain language statement expla consent form, both approved by the Human Ethics Advisory Group from the U instructed to take the session in a quiet place with uniform lighting and a neutr interruptions and potential bias. Besides the written instructions, a video was p the session explaining these instructions with subtitles for non-native English s participant was verified before the researcher started the sessions to initiate the The labels were randomized once (prior to the questionnaire development) and participants; the order presented is shown in Figure 1. The questionnaire cons acceptability based on different attributes, such as perceived strength, pleasant FaceScale (emotional response), perceived quality, and willingness to purchas apply (CATA) questions for emojis and preferred areas of interest from label e

CATA Emojis
, the participants were asked to read the plain language statement explaining the experiment and sign a form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were also ed to take the session in a quiet place with uniform lighting and a neutral background to avoid any tions and potential bias. Besides the written instructions, a video was presented to each participant before ion explaining these instructions with subtitles for non-native English speakers. The testing setting of each ant was verified before the researcher started the sessions to initiate the test. els were randomized once (prior to the questionnaire development) and presented in a fixed order for all ants; the order presented is shown in Figure 1.

CATA Emojis
, biometrics through facial recognition to further assess their emotional subcons session, the participants were asked to read the plain language statement expla consent form, both approved by the Human Ethics Advisory Group from the U instructed to take the session in a quiet place with uniform lighting and a neutr interruptions and potential bias. Besides the written instructions, a video was p the session explaining these instructions with subtitles for non-native English s participant was verified before the researcher started the sessions to initiate the The labels were randomized once (prior to the questionnaire development) and participants; the order presented is shown in Figure 1. The questionnaire cons acceptability based on different attributes, such as perceived strength, pleasant FaceScale (emotional response), perceived quality, and willingness to purchas apply (CATA) questions for emojis and preferred areas of interest from label e

CATA Emojis
, were conducted via Zoom (Zoom Video Communications, Inc., San Jose, CA, USA) using Google F LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host t session in the background and record videos of the participants while looking at the label (15 s) to o biometrics through facial recognition to further assess their emotional subconscious responses [16,1 session, the participants were asked to read the plain language statement explaining the experiment a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). instructed to take the session in a quiet place with uniform lighting and a neutral background to avoi interruptions and potential bias. Besides the written instructions, a video was presented to each parti the session explaining these instructions with subtitles for non-native English speakers. The testing s participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed o participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessmen acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], ar FaceScale (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2). ountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor the in the background and record videos of the participants while looking at the label (15 s) to obtain their ics through facial recognition to further assess their emotional subconscious responses [16,19]. Before the the participants were asked to read the plain language statement explaining the experiment and sign a form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were also ed to take the session in a quiet place with uniform lighting and a neutral background to avoid any tions and potential bias. Besides the written instructions, a video was presented to each participant before ion explaining these instructions with subtitles for non-native English speakers. The testing setting of each ant was verified before the researcher started the sessions to initiate the test. els were randomized once (prior to the questionnaire development) and presented in a fixed order for all ants; the order presented is shown in Figure 1.

CATA Emojis
, and shared the emojis Beverages 2021, 7, x FOR PEER REVIEW conducted using SAS ® v. 9.4 (SAS Institute, Cary, NC, USA), the number of p to find significant differences between samples (1 − β > 0.99). The sessions las were conducted via Zoom (Zoom Video Communications, Inc., San Jose, CA, LLC, Mountain View, CA, USA) to display the questionnaire and labels. This session in the background and record videos of the participants while looking a biometrics through facial recognition to further assess their emotional subcons session, the participants were asked to read the plain language statement expla consent form, both approved by the Human Ethics Advisory Group from the U instructed to take the session in a quiet place with uniform lighting and a neutr interruptions and potential bias. Besides the written instructions, a video was p the session explaining these instructions with subtitles for non-native English s participant was verified before the researcher started the sessions to initiate the The labels were randomized once (prior to the questionnaire development) and participants; the order presented is shown in Figure 1. The questionnaire cons acceptability based on different attributes, such as perceived strength, pleasant FaceScale (emotional response), perceived quality, and willingness to purchas apply (CATA) questions for emojis and preferred areas of interest from label e ed using SAS ® v. 9.4 (SAS Institute, Cary, NC, USA), the number of participants was statistically sufficient ignificant differences between samples (1 − β > 0.99). The sessions lasted 10-15 min per participant and nducted via Zoom (Zoom Video Communications, Inc., San Jose, CA, USA) using Google Forms (Google, ountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor the in the background and record videos of the participants while looking at the label (15 s) to obtain their ics through facial recognition to further assess their emotional subconscious responses [16,19]. Before the the participants were asked to read the plain language statement explaining the experiment and sign a form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were also ed to take the session in a quiet place with uniform lighting and a neutral background to avoid any tions and potential bias. Besides the written instructions, a video was presented to each participant before ion explaining these instructions with subtitles for non-native English speakers. The testing setting of each ant was verified before the researcher started the sessions to initiate the test. els were randomized once (prior to the questionnaire development) and presented in a fixed order for all ants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of bility based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,22], le (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check all that ATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

Video Analysis to Obtain Biometrics
All videos from participants recorded per sample were analyzed in batch using an automatic comput developed by the Digital Agriculture Food and Wine (DAFW) group based on the Affectiva (Affect MA, USA) software development kit (SDK). This application is able to assess participants' facial ex videos using a histogram of the oriented gradient algorithms, which are automatically translated into related emojis (Table 3) using support vector machine modeling algorithms in a batch analysis fashi

Statistical Analysis
All quantitative data (Tables 1 and 3) were analyzed through ANOVA to assess significant differen

CATA Emojis
was most selected for fun and natural labels.  Table 6 shows non-significant differences (p < 0.05) between samples for the selection of logo, intensity scale, and product name. In premium, the most selected areas of interest (AOI) were colors, layout, and pattern/texture, showing significant differences with fun and classic. The latter was the lowest in the frequency of selections for all AOI. LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor the session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before the session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were also instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant before the session explaining these instructions with subtitles for non-native English speakers. The testing setting of each participant was verified before the researcher started the sessions to initiate the test.

Multivariate Data Analysis
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for all participants; the order presented is shown in Figure 1.       , and sadness. The everyday label concept was associated with the selection of colors as the best AOI; biometric responses, such as joy and smile; self-reported responses, such as pleasantness; and a selection of emojis, such as interruptions and potential bias. Besides the written instructions, a video was p the session explaining these instructions with subtitles for non-native English s participant was verified before the researcher started the sessions to initiate the The labels were randomized once (prior to the questionnaire development) and participants; the order presented is shown in Figure 1. The questionnaire cons acceptability based on different attributes, such as perceived strength, pleasant FaceScale (emotional response), perceived quality, and willingness to purchas apply (CATA) questions for emojis and preferred areas of interest from label e

CATA Emojis
, the session explaining these instructions with subtitles for non-native English speakers. The testing s participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed o participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessmen acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], ar FaceScale (emotional response), perceived quality, and willingness to purchase (Table 1), as well as apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
, and ed to take the session in a quiet place with uniform lighting and a neutral background to avoid any tions and potential bias. Besides the written instructions, a video was presented to each participant before ion explaining these instructions with subtitles for non-native English speakers. The testing setting of each ant was verified before the researcher started the sessions to initiate the test. els were randomized once (prior to the questionnaire development) and presented in a fixed order for all ants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of bility based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,22], le (emotional response), perceived quality, and willingness to purchase (Table 1), as well as check all that ATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
. Furthermore, the premium label was more associated with the self-reported responses, such as perceived strength; the selection of brand as the best AOI; PQuality; a selection of emojis, such as biometrics through facial recognition to further assess their emotional subconscious responses [16,1 session, the participants were asked to read the plain language statement explaining the experiment a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). instructed to take the session in a quiet place with uniform lighting and a neutral background to avoi interruptions and potential bias. Besides the written instructions, a video was presented to each parti the session explaining these instructions with subtitles for non-native English speakers. The testing s participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed o participants; the order presented is shown in Figure 1. The questionnaire consisted of the assessmen acceptability based on different attributes, such as perceived strength, pleasantness (valence) [20], ar FaceScale (emotional response), perceived quality, and willingness to purchase (Table 1), as well as apply (CATA) questions for emojis and preferred areas of interest from label elements ( Table 2).

CATA Emojis
and conducted using SAS ® v. 9.4 (SAS Institute, Cary, NC, USA), the number of p to find significant differences between samples (1 − β > 0.99). The sessions las were conducted via Zoom (Zoom Video Communications, Inc., San Jose, CA, LLC, Mountain View, CA, USA) to display the questionnaire and labels. This session in the background and record videos of the participants while looking a biometrics through facial recognition to further assess their emotional subcons session, the participants were asked to read the plain language statement expla consent form, both approved by the Human Ethics Advisory Group from the U instructed to take the session in a quiet place with uniform lighting and a neutr interruptions and potential bias. Besides the written instructions, a video was p the session explaining these instructions with subtitles for non-native English s participant was verified before the researcher started the sessions to initiate the The labels were randomized once (prior to the questionnaire development) and participants; the order presented is shown in Figure 1. The questionnaire cons acceptability based on different attributes, such as perceived strength, pleasant FaceScale (emotional response), perceived quality, and willingness to purchas apply (CATA) questions for emojis and preferred areas of interest from label e

CATA Emojis
; and biometric responses, such as relaxed and valence. On the other hand, the classic label was associated with the self-reported responses for the selection of product names as the best AOI and most neutral to negative emojis, such as conducted using SAS ® v. 9.4 (SAS Institute, Cary, NC, USA), the number of participants was statistically sufficient to find significant differences between samples (1 − β > 0.99). The sessions lasted 10-15 min per participant and were conducted via Zoom (Zoom Video Communications, Inc., San Jose, CA, USA) using Google Forms (Google, LLC, Mountain View, CA, USA) to display the questionnaire and labels. This setup allowed a host to monitor the session in the background and record videos of the participants while looking at the label (15 s) to obtain their biometrics through facial recognition to further assess their emotional subconscious responses [16,19]. Before the session, the participants were asked to read the plain language statement explaining the experiment and sign a consent form, both approved by the Human Ethics Advisory Group from the UoM (ID: 1953926.4). They were also instructed to take the session in a quiet place with uniform lighting and a neutral background to avoid any interruptions and potential bias. Besides the written instructions, a video was presented to each participant before the session explaining these instructions with subtitles for non-native English speakers. The testing setting of each participant was verified before the researcher started the sessions to initiate the test.
The labels were randomized once (prior to the questionnaire development) and presented in a fixed order for all participants; the order presented is shown in Figure 1.

CATA Emojis
. The natural and fun labels were located closer to the MFA center, with the former being located on the positive emotion side of the graph and the latter presenting negative associations with the bold label's characteristics. ATA) questions for emojis and preferred areas of interest from label elements ( Table 2). ), perceived quality, and willingness to purchase ( Table 1), as well as check all that ojis and preferred areas of interest from label elements ( Table 2).
ributes assessed for acceptability of the coffee pod label samples. ants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of bility based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,22], le (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check all that ATA) questions for emojis and preferred areas of interest from label elements ( Table 2). nce (prior to the questionnaire development) and presented in a fixed order for all ed is shown in Figure 1. The questionnaire consisted of the assessment of nt attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,22], e), perceived quality, and willingness to purchase ( Table 1), as well as check all that mojis and preferred areas of interest from label elements ( Table 2).
tributes assessed for acceptability of the coffee pod label samples.  (Table 1), as well as check all that reas of interest from label elements ( Table 2). acceptability of the coffee pod label samples.

Scale Anchors
9-point scale 1: Extremely mild-9: Extremely strong int hedonic scale 1: Unpleasant-9: Pleasant 9-point scale 1: Relaxed-9: Stimulated int hedonic scale 9-point scale 1: Extremely low-9: Extremely high 9-point scale 1: Extremely unlikely-9: Extremely likely pply (CATA) questions for assessment of the coffee pod label samples. NS els were randomized once (prior to the questionnaire development) and presented in a fixed order for all ants; the order presented is shown in Figure 1. The questionnaire consisted of the assessment of ility based on different attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,22], le (emotional response), perceived quality, and willingness to purchase ( Table 1), as well as check all that ATA) questions for emojis and preferred areas of interest from label elements ( Table 2). nce (prior to the questionnaire development) and presented in a fixed order for all ed is shown in Figure 1. The questionnaire consisted of the assessment of nt attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,22], e), perceived quality, and willingness to purchase (Table 1), as well as check all that mojis and preferred areas of interest from label elements ( Table 2).

ATA Emojis
tributes assessed for acceptability of the coffee pod label samples.  ons and potential bias. Besides the written instructions, a video was presented to each participant before n explaining these instructions with subtitles for non-native English speakers. The testing setting of each nt was verified before the researcher started the sessions to initiate the test.
ls were randomized once (prior to the questionnaire development) and presented in a fixed order for all nts; the order presented is shown in Figure 1. a quiet place with uniform lighting and a neutral background to avoid any . Besides the written instructions, a video was presented to each participant before tructions with subtitles for non-native English speakers. The testing setting of each the researcher started the sessions to initiate the test.
ce (prior to the questionnaire development) and presented in a fixed order for all d is shown in Figure 1. The questionnaire consisted of the assessment of t attributes, such as perceived strength, pleasantness (valence) [20], arousal [21,22], ), perceived quality, and willingness to purchase (Table 1), as well as check all that ojis and preferred areas of interest from label elements ( Table 2).
ributes assessed for acceptability of the coffee pod label samples.

Video Analysis to Obtain Biometrics
All videos from participants recorded per sample were analyzed in batch using an automatic computer application developed by the Digital Agriculture Food and Wine (DAFW) group based on the Affectiva (Affectiva, Boston, MA, USA) software development kit (SDK). This application is able to assess participants' facial expressions from videos using a histogram of the oriented gradient algorithms, which are automatically translated into emotions and related emojis ( Table 3)     located on the positive emotion side of the graph and the latter presenting negative associations with the bold label's characteristics. Figure 2b shows the matrix with significant correlations between the self-reported acceptability and biometric responses. It can be observed that, despite being low, there were negative and significant correlations (p < 0.05) between and pleasantness (r = −0.16), PQuality (r = −0.17), FaceScale (r = −0.12), and WPurchase (r = −0.17). Similar correlations were found between anger facial expression and the aforementioned self-reported responses. Furthermore, there was a negative low but significant correlation between surprise and FaceScale (r = −0.12), and positive correlations between self-reported arousal and engagement and smile (r = 0.11). On the other hand, PQuality was positively correlated with strength (r = 0.63), pleasantness (r = 0.71) and FaceScale (r = 0.79).
(a)  Table 1; CATA: check all that apply; AOI: area of interest; F1 and F2: factors one and two, respectively.

Virtual Sensory Sessions
The virtual sensory sessions for individual participants ran smoothly, and data gathering and analysis were completed within three weeks, with only one person required for data collection and one for data analysis. The time of virtual sessions and data analysis can be reduced by having multiple participants in one session through Zoom, in which they can all be added to a single screen that is recorded. New video analysis algorithms developed by the DAFW group can crop each participant, thereby allowing them to be analyzed automatically through the batch code described in this study using Affectiva algorithms. An advantage of sensory tests during social isolation, using the available digital tools, is that potential panelists are in familiar environments, such as their homes, places dedicated to work, or where social interactions may occur when doing the test with family members or friends [24]. In places dedicated to work, consumers are more likely to test different products. To date, there are no published papers using or proposing virtual sensory sessions that include biometrics recording to be used in emergency situations, such as extensive lockdown during pandemics, in which participants may be presented with conditions that do not allow them to attend a sensory laboratory or to reach more participants in other cities or countries.

Emotional Responses from Biometrics and Emoji Selection
The most accepted label from this study with positive emotional responses and positive emojis was the premium concept, which is in accordance with other label studies for chocolates [4]. The opposite responses were obtained for the bold label with the cheese dip concept, which could have been associated with cheesy aroma in the coffee (Tables 4  and 5). It has been shown that smell and positive aroma descriptors are determinants of  Table 1; CATA: check all that apply; AOI: area of interest; F1 and F2: factors one and two, respectively.

Virtual Sensory Sessions
The virtual sensory sessions for individual participants ran smoothly, and data gathering and analysis were completed within three weeks, with only one person required for data collection and one for data analysis. The time of virtual sessions and data analysis can be reduced by having multiple participants in one session through Zoom, in which they can all be added to a single screen that is recorded. New video analysis algorithms developed by the DAFW group can crop each participant, thereby allowing them to be analyzed automatically through the batch code described in this study using Affectiva algorithms. An advantage of sensory tests during social isolation, using the available digital tools, is that potential panelists are in familiar environments, such as their homes, places dedicated to work, or where social interactions may occur when doing the test with family members or friends [24]. In places dedicated to work, consumers are more likely to test different products. To date, there are no published papers using or proposing virtual sensory sessions that include biometrics recording to be used in emergency situations, such as extensive lockdown during pandemics, in which participants may be presented with conditions that do not allow them to attend a sensory laboratory or to reach more participants in other cities or countries.

Emotional Responses from Biometrics and Emoji Selection
The most accepted label from this study with positive emotional responses and positive emojis was the premium concept, which is in accordance with other label studies for chocolates [4]. The opposite responses were obtained for the bold label with the cheese dip concept, which could have been associated with cheesy aroma in the coffee (Tables 4 and 5). It has been shown that smell and positive aroma descriptors are determinants of consumer perception [15], which could explain this study's results. Furthermore, the second-highest scores were for the natural concept, which incorporated grains of coffee and more earthy colors associated with a healthy option in the imagery, which is also in accordance with other studies involving healthy and organic concepts within labels [13,25,26].

Analysis of Areas of Interest within Labels
Following the results from the emotional and emoji responses, the premium label was highlighted with statistically higher areas of interest selection, such as colors, layout, and pattern/texture, with layout being the highest area of interest for the natural concept and color for the bold concept (Table 6). However, analysis of features was limited in this study; more specific assessments can be conducting when analyzing label features using eye-trackers [2,3,[27][28][29] or more practically for virtual sensory sessions when using digital tools (from video analysis) for eye tracking to assess fixation number and duration that the panelists spend at each component [30,31]. The DAFW group has developed algorithms to analyze eye-tracking parameters and the emotional response based on facial expressions from the different features of labels [32]. Hence, further studies using the methodologies proposed here will enrich the number of parameters to be analyzed and find different patterns within the data or apply machine learning algorithms for artificial intelligence application to label analysis.

Multivariate Data Analysis
The multivariate data analysis (Figure 2a), including the emotional response, emoji selection, and analysis of label areas of interest, offered a more comprehensive analysis of the relationships and patterns among the different labels and concepts. The main addressable features with this analysis are the positive selection of graphics with the least preferred concept (bold), which shows that even if the graphics are preferred, the concepts within the label related to the specific product are more important for consumers. Similar assessment can be conducted for the concept of fun, which was closely related to the panelists' color interests. Furthermore, it was also found that the name was mostly associated with positive and negative emojis, with the classic concept offering more interest but without correlated results for preference.
In the case of the correlations among the self-reported and subconscious responses (Figure 2b), it was found that consumers rate labels as higher quality when they perceive coffee as higher strength, when labels are more pleasant, and when they elicit more positive emotions; the latter is consistent with findings from Gunaratne et al. [2] for chocolate label assessment. Furthermore, the willingness to purchase is directly influenced by the perceived quality, as found in other research [33].
One of the limitations of this study with the methodology proposed was the requirement of testing one participant at a time due to video recording using the available meeting software capabilities. However, this was solved after the study was conducted by assessing a video of multiple participants in a single screen recording and automatically cropping participants for further analysis. This was achieved by coding video cropping using computer vision algorithms in Matlab ® . Furthermore, this study was based on online easy-to-access and free tools, such as Zoom (free sessions for up to 1 h), Google Forms, and Affectiva SDK. More information can be accessed using more specific self-reported software packages specific for sensory analysis, such as RedJade [34,35], Compusense [36,37], and Fizz [38], among others. However, these packages may be cost prohibitive for small and medium food and beverage companies.
Further research should be conducted using more specialized software to obtain more information that can be relevant to companies' decision making regarding product and package development.

Conclusions
Social isolation during the pandemic of 2020 served as an incentive for many software companies to develop virtual or remote sensory tests to comply with social distancing Beverages 2021, 7, 5 13 of 15 and lockdowns. This study implemented new and emerging sensory tools to analyze coffee labels, thereby obtaining results consistent with previous studies for coffee and other beverage labels using normal sensory analysis with panelists attending physically sensory sessions. This research showed that free online software resources, such as remote meeting applications, video capture, and self-reported forms, are effective tools to carry out sensory analysis of labels outside the laboratory when social isolation and distancing are required, with results comparable to other more established methods. These techniques can also be applied to other food and beverage products by sending them to consumers via courier/mail to be tested using sensory techniques. The tools proposed in this study are free and open-source software tools, such as Google Forms, Affectiva, and Zoom, the latter with free usage for multiple 1 h sessions. These software packages are advantageous to conduct label and even food and beverage product sensory analysis compared to commercial software that can do similar sensory studies as described in this paper by incorporating facial expressions and emotional responses. However, they are cost-prohibitive, especially for small and medium food and beverage companies. Furthermore, this study proposed more efficient ways to conduct these sessions with multiple participants on single videos and incorporate eye-tracking software from videos to analyze the emotional response to specific features from different beverage products' labels. The system proposed in this study could be of great benefit to food and beverage companies not only in the context of isolation conditions but by increasing the reach of the sensory trials to other countries, thereby incorporating a higher number of participants unrestricted by sensory laboratory space, different cultural backgrounds, or age and without international or language boundaries. Supplementary Materials: The following are available online at https://www.mdpi.com/2306-571 0/7/1/5/s1, Table S1. Mean ± standard error values from the biometric emotional responses for the six different label samples (Figure 1).

Informed Consent Statement:
All subjects involved in the study signed a consent form prior to the sensory session.
Data Availability Statement: Data and intellectual property belong to The University of Melbourne; any sharing needs to be evaluated and approved by the University.