How Commercial Food Videos Affect Female Customers

Measuring Female Bio-response Towards Commercial Food Videos

Xintong Zhu, Yong Wang, ZhenZheng Wang, Xiguang Wang and Chen Wang

Future Media Convengence Institute, Xinhua Net, Jinyu Building, Xicheng District, Peking, China

Keywords: Physiological Computing, GSR Sensors, Audience Experiences, Advertising Videos.

Abstract: The concept design of a food television commercial (TVC) could affect the effectiveness of immersive media

and user experience. Traditional methods (e.g., surveys or eye tracking) for evaluating the targeted consumer’s

responses are serious limited in several aspects. In this paper, through closely working with the TVC

designers, we used the data gathered from physiological sensor to measure viewers’ watching experiences.

We thereby conclude how we used our own Galvanic Skin Response (GSR) sensors to measure audience

responses to the three clips of food TVCs. The results demonstrate that GSR sensors can provide fine-grained

information for the advertisement community. Compared to subjective evaluation methods, the continuous

user experience can be vividly visualized, and this enables designers to efficiently evaluate the impact of a

TVC.

1 INTRODUCTION

The advertisement industry has been fully aware that

a content design for a commercial campaign plays a

vital role in spreading the name and the impact of

products. Different food promotion strategies shall be

specified with different concept designs in a

campaign. In particular, to attract the targeted

customers, advertising companies always seek for the

optimal way to design a television commercial (TVC)

that could seize the attention and further bringing the

tangible revenues.

However, it is a great challenge to measure the

impact of a TVC. First, obtaining such information is

rather expensive and time consuming. For instance,

the commercial companies generally hire agencies to

run market researches that normally take several

months. One of the feasible approaches is to carry out

online questionnaires to collect the answers from the

target customers and analyse the data. Second,

traditional methods (e.g., questionnaires) are not the

most appropriate one to measure the impact of a TVC,

since questionnaire data is discrete, which means the

data is hard to be connected to the whole duration of

a TVC. In addition, subjective reports can easily

collect information from the questions that users are

aware of, but it is unlikely to ask them to answer some

questions related to attention or engagement (Latulipe

et al., 2011). Last, recently, eye movements

Figure 1: The experimental session.

(Krugman et al., 1994)and facial expressions (Joho et

al., 2009)have risen as popular methods to measure

the impact of a TVC. We doubt the credibility of such

methods as a watcher might be completely absent-

minded when they are watching a video. If so, eye

tracking and facial expression data seem to be

particularly meaningless for evaluating the influence

of a TVC, since the relationship between user inner

attention and visual attention still remains unclear.

Physiological measurements have many

advantages as an alternative tool to monitor, in real-

time, audience responses in a video consumption. The

physiological data are continuous, providing time

series that can be mapped to events (e.g., the moment

that a brand logo appears) of a TVC. Second, the non-

intrusive nature of the physiological measurements

guarantees a seamless experience of the audience. In

Zhu, X., Wang, Y., Wang, Z., Wang, X. and Wang, C.

How Commercial Food Videos Affect Female Customers - Measuring Female Bio-response Towards Commercial Food Videos.

DOI: 10.5220/0006394300350044

In Proceedings of the 4th International Conference on Physiological Computing Systems (PhyCS 2017), pages 35-44

ISBN: 978-989-758-268-4

addition, physiological sensors have been applied in

audience research, e.g., user emotion (Oliveira et al.,

2011)and user engagement(Picard and Daily, 2005).

Although some studies use them to assess the impact

of a video (Yoon et al., 1998), the audio effect

(Kenning and Plassmann, 2008), or use question-

naires to investigate audience’s purchase behaviour.

Surprisingly, no studies have ever used physiological

sensors to measure women’s experience when

watching a commercial. None of those studies has

paid particular attention to female customers, who are

most of the time the actual decision maker in a

household’s shopping journey.

In this paper, requested by the producers who

have rich experiences in designing commercial

videos, we dive into female customers’ experience in

commercial video consumption. Specifically, we first

interviewed the producers and collected their research

interests. Then, we designed the experiment and

chose the appropriate methods to answer the research

questions. In our case, we chose GSR sensors to

measure user experience, as they are highly accurate

for indicating the user arousal and the emotional

intensities compared to other bio sensors. Taking

advantage of the sensor data, we could map their

response to specific events (e.g., the moment that the

food brand appears).

Our contribution has twofold. On the one hand,

we focus our research interests on female customers,

since their watching experience has never been

investigated in the previous studies. On the other

hand, we use GSR sensors, instead of questionnaires,

to continuously record user responses in a food TVC,

a research area that has not been explored before.

1.1 Research Questions

The research questions were generalized from the

interviews with the producers. They are both very

experienced in commercial video production. They

believe that there are many ways to produce a good

TVC, but how to demonstrate that one concept design

is superior than the other one is a challenge. In

particular, the producers consider that how female

customers are influenced by these TVCs is extremely

important, because they are major buyers in

households. In other words, if a TVC hits their heart,

the chance is high that they will buy the product in the

future. The producers proposed three TVCs to run the

experiments, in which the different types of food are

involved. They had particular interests in female

reactions, i.e., how female would react when the food

close-up shot first appears, how do they response

differently when watching the three videos, and how

their bio-responses are related to their subjective

reports.

The interview was semi-conducted and coded

based on the notes. Four researchers participated and

concluded three research questions:

R1: What are the differences in female consumers’

responses when watching the three videos?

R2: How do the female consumers react to the close-

up of food first appearing in the videos?

R3: How is the female bio-response correlated to the

subjective reports?

The remainder of this paper is organized as

follows. Related work will be reported first and the

design alternatives for the hardware and the software

are introduced next. Then the results are exhibited.

followed with the discussion and the conclusion at

last.

2 RELATED WORK

There are various definitions for audience response.

For instance, some studies use different synonyms,

such as audience engagement, audience feedback,

audience interests, audience interaction and audience

(or user) experience. Christopher Peters et al. (Peters

et al., 2009) described audience response as a

combination of focus, interest, perception, cognition,

experience and action. On the other hand, Heather L.

O’Brien et al. (O'Brien and Mclean, 2009) pointed out

that audience engagement (response) can facilitate

users with more enriching interactions in computer

applications: engaged users tend to recommend the

products (or service) to others. In addition, in

affective computing, users’ emotional response

(Wang, Prendinger, and Igarashi, 2004) is obtained as

an evaluation tool to define user engagement. In game

application studies (Fischer and Benford, 2009),

audience engagement (response) refers to players’

state awareness and synchronization. Audience

biofeedback, e.g., arousal, is also used as an indicator

of the levels of players’ engagement(Chanel et al.,

2008).

Audience response can be measured in two ways:

explicitly and implicitly. Explicit methods normally

require users’ intentional inputs, like surveys or

ratings, whereas implicit measurements generally

capture audience feedback through physiological

sensors, such as GSR sensors. During the process of

data collection, sensor readings are collected in real

time without interrupting audience’s watching

journey.

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

Visual analysis methods, e.g., eye movements, are

also classified into the scope of explicit methods of

measuring audience response. In such studies, eye-

gaze, eye movements and head movement trackers

are installed to define users’ interested spots during a

video consumption or other applications like game

studies. Some other methods, e.g., physiological

sensors, are also combined with visual analysis

methods to capture audience response, in which

audience inner state are characterized, e.g., boredom

or fatigue. However, these studies could not avoid

intentional requiring audience annotations constantly,

by which the labels of the attributes of the data are

generated(Kapoor et al., 2007).

Audience response has a significant impact on

many applications. There are plenty of quantitative

studies executed in game researches based on the

reliability and suitability of physiological sensors.

Pejam et al.(Mirza-Babaei et al., 2013) have explored

the possibilities of improving games design by

providing user biofeedback, and the results showed

that a combination of user GSR feedback would help

designers choose proper design strategy for higher

game play quality. In the game industries, Sony, for

example, recently has decided to add GSR sensor into

their new game controller, DualShock 4 (DualShock,

2017), where users’ GSR-response is detected as

interest level towards the game. Another example

(Sakurazawa et al., 2003) is that users’ GSR-response

is used as the users’ agitation in a game, and the more

often a user feels agitated, the more enemies will

appear in a game.

Previous studies have also shown that GSR sensor

is one of the indicators for users’ cognitive and

emotional status. Lin et al (Lin et al., 2008)

successfully investigated how audience GSR-

response differs in the different movie sections. In

particular, the fluctuations of GSR data were linked

to events during a 3D movie experience.

Matthew et al. (Pan et al., 2011)demonstrated a

novel interaction by using GSR sensors in an audio

stream bookmarking, where users’ GSR-response

was monitored as a response to the external

interruptions. GSR sensors were also applied in a

wearable system in order to help users select the high

arousal photos, which were the most relevant to the

users’ ordinary daily life(Sas et al., 2013).

Web applications are also benefited from

audience response research. In the paper(Lunn et al.,

2010), users’ biofeedback was used to distinguish

audience response within the different age scopes, in

terms of a web 2.0 application. Audience affective

states were also employed to investigate what kind of

interaction technique on the web has a significant

Figure 2: The GSR sensors and the monitoring system.

impact on users(Hart et al., 2012). Transforming

audience physiological signals (audience response)

into a smile icon was implemented in an online

chat(Wang et al., 2004). Users’ biofeedback on

preference, e.g., like or dislike, was used as the input

data in order to improve the accuracy of online

recommendation system(Madan et al., 2004).

Audience response also plays an important role in

some other applications. For instance， in Olympic

Games, audience clapping frequency was visualized

on the display screen to encourage athletes’

performance(Barkhuus, 2008); audience ‘cheering

meter” was measured to aid voting at rap

competitions(Aigner et al., 2004).

GSR, is also known as galvanic skin response,

electro dermal response (EDR), psych galvanic reflex

(PGR), skin conductance response (SCR), or skin

conductance level (SCL). GSR sensors measure the

users’ electrical conductance of the skin, where users’

sweat glands are varied and controlled by the

sympathetic nervous system. Therefore, GSR sensors

are normally considered as an indicator of

psychological or physiological arousal or stress.

When users are highly aroused, users’ skin

conductance increases in turn. GSR sensors can be

either purchased from commercial companies or self-

developed. Commercial companies, such as BioPac,

Thought technology and Q sensors, offer this type of

GSR sensor with a high price. Although such

commercial sensors allow researchers to start

experimenting immediately, they do not provide

functions to measure groups of users simultaneously.

The reason is that the communication protocol

normally is Bluetooth, which has limitation to

connect cell nodes in wireless network, e.g., a master

and up to 7 slave piconet networks.

3 METHODOLOGY

3.1 Participants

All the participants were recruited from our institute

per the requirements (from 25 to 34) of the producers,

How Commercial Food Videos Affect Female Customers - Measuring Female Bio-response Towards Commercial Food Videos

and they had not any visual or acoustic problems

(Figure 1). All participants were divided into 4 teams

with 4 persons in each.

3.2 Stimuli and Apparatus

The stimuli consisted of 3 different short videos

(video A, B, and C). There was a short pause left

between video clips (32 seconds) with grey screen to

let users jump out of the previous watching

experience. The video A is intended to prompt the

organic blueberry, the video B advertises the ham

product, and the video C addresses on the beer

chicken. The three videos were played in different

sequence for each team in order to minimize the

influence of order. The duration of the experiments

was 12 minutes 34 seconds in total.

We built the GSR sensors using a Jeenode board

with a RF12 wireless module, a low pass filter and

some accessories (e.g., electrodes) (Figure 2). The

wireless function of the RF12 module makes it

possible for simultaneously measuring audience at a

large scale.

The sensors have been validated through a

number of experiments (anonymous). All the sensor

subordinates simultaneously send GSR data packets

back

to the master sink node, which is connected to a

laptop. The master node communicates with all the

subordinate nodes by using a polling mechanism. In

our case, we set up the sampling rate at 2Hz, which is

not optimal, but it is sufficient to recover the sensor

data. The sensors are robust against noise because of

the circuit design and the data smoothing procedure

in the software (Figure 3). In addition, all the sensor

data were synchronized with the time stamp of the

videos.

Figure 3: the diagram of the feature extraction of the GSR

data.

3.3 Interviews and Questionnaires

The purpose of the interviews and questionnaires is to

better interpret the sensor data. All participants were

interviewed before and after they watched the videos.

The interviews mainly focused on the following five

aspects: the title of their job, their favourite food, their

favourite video，which specific scenes in the videos

that they were interested in, and reasons for likes and

dislikes.

The questionnaires were designed at seven scales.

The pre-questionnaires include the following

questions:

1. What is your job and responsibilities?

2. what kind of food do you like to eat?

3. what kind of food you do not like to eat?

4. what taste of food do you like?

5. what taste of food you do not like?

6. what kind of video do you like?

7. what kind of video you do not like?

8. Do you often watch video by smart phone?

9. Do you often buy food online?

The post-questionnaires have three questions:

1. What do you want to eat after watching?

2. Which video do you think the best?

3. Which style of the video do you like? And the

reason?

3.4 Software

All the videos were played on the four same-type

smart phones, and the participant hold the phone in

their most comfortable gesture. The whole

experiment was video-recorded by using the facilities

installed in our user lab (Figure 4). The recording

video streams were collected by the software written

in Python, which is installed in the observation room.

All the sensor data was analysed by SPSS and

MathWorks (Matlab).

3.5 Experimental Procedures

Before the experiment started, all the participants

were asked to fill an informed consent form and the

pre-questionnaires. Then the interview began and oral

instructions were provided. After that, the

experimenters helped the participants to wear the

sensors, and then they opened the video play

program, and clicked the “play” button. When the

experiment was finished, the interviews and the post-

questionnaires were conducted after the sensors were

taken off.

3.6 Data Analysis

To explore the participant’s GSR response, both the

event-related skin conductance response (SCR) and

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

Figure 4: the user lab and the observation room.

the skin conductance level (SCL) were analysed and

normalized by Z-score. In addition to that, we also

used analysis of variance (ANOVA) and Least

Significant Difference (LSD) to test the significance

of physiological data in three videos.

There are several steps to analyse Electro Dermal

Activity (EDA) to extract the event-related SCR data

based on Fleureau (

Fleureau, Julien, Philippe Guillotel,

and Izabela Orlac. 2013

) and his colleagues’ work.

Normally, when participants receive an engaging

stimulus, their GSR value will increase quickly with

a latency of 1-3 seconds, and after reaching a

maximum value, it will recover to a value around the

baseline level. Firstly, a 2Hz (G(t)) low-pass filter

was applied to remove noise from the raw data, such

as other physiological signals and electrical noise.

Then a derivation (from G(t) to G’(t) ) was applied to

calculate the rate of change of the GSR data. In doing

so, we could know if the GSR value is ascending

(positive values) or descending (negative values).

After that all the negative values were discarded, with

only the positive ones being kept (from G’(t) to

G’

(t)), which means that we only focus on the

increasing phases of the GSR signal, because the

negative phases only reflect the recovery of the signal

to the baseline. These steps helped us to extract the

SCR data from the raw electrophysiological data.

(1)

To temporally analyse emotional flow, we

applied an overlapping time moving window with a

window size of 30 samples (30 seconds), and an

overlap of 15 samples (15 seconds). This step helped

us to smooth the data and remove the users’ GSR

latency. So the mean values of G’

(t) were converted

into G(i) (1 ≤ i ≤ k, k is the number of the moving

windows). G(i) is the mean derivative value of one

subsample in one specific moving window.

Since each individual may have a different

amplitude for the derivative GSR signal when

exposed to the same stimulus, G(i) was divided by the

sum of the subsampled skin response values (Formula

(1)), and the output was G

(i) (1 ≤ n ≤ N, N is the

number of the sample; 1 ≤ i ≤ k, k is the number of the

moving windows).

(i) is the individual value in a moving window,

which cannot represent the whole group’s response,

considering some outliers, differences from person to

person, noise (e.g. body movements) and so on. To

define whether the group had a significant arousal or

not, a statistical test called the bilateral Mann-

Whitney-Wilcoxon (MWW) test was used. This test

detects whether there is a significant difference

between the audience arousal response (ܩ

(i) and the

background noise. We took the lowest 10% of the

values in ܩ

(i) as background noise. ܩ

(i) of a single

time sample was compared to the background noise

of each time sample, which means that we used

MWW test to compare k times and obtain k p-values

for each time sample. The final p-value of each time

sample is the averaged value of those k p-values. For

final p-values lower than 5%, we considered the

response during that time sample to be significantly

different from the background noise.

The mean value of EDA of the first 32s blank

video (grey screen) were used as the baseline, which

was then subtracted from the raw

electrophysiological data, to remove individual

differences.

Z-score is the number of standard deviations

from the mean a data point is. Z-scores range from -

3 standard deviations (which would fall to the far

left of the normal distribution curve) up to +3

standard deviations (which would fall to the far right

of the normal distribution curve). The basic z score

formula for a sample is:

z = (x – μ) / σ (2)

where, μ is the mean of the population, σ is the

standard deviation of the population. The absolute

value of z represents the distance between the raw

score and the population mean in units of the standard

deviation. z is negative when the raw score is below

the mean, positive when above.

Here, we use Z-score to standardize the EDA in

order to compare the changes of EDA when different

short videos were played.

The ANOVA is used to determine whether there

are any statistically significant differences between

the means of three or more independent (unrelated)

groups. When ANOVA gives a significant result, this

How Commercial Food Videos Affect Female Customers - Measuring Female Bio-response Towards Commercial Food Videos

indicates that at least one group differs from the other

group. Yet, the omnibus test does not indicate which

group differs. In order to analyse the pattern of

difference between means, the ANOVA is often

followed by specific comparisons, and the most

commonly used method is to compare two means.

Here, we used least significant difference (LSD) to

compute the smallest significant difference between

two means as if these means had been the only means

to be compared (i.e., with a t test) and to declare any

significant difference larger than the LSD. So, we

firstly used ANOVA to analyse weather the EDA is

different between some close-up shot and normal

frame, and then the LSD analysis was used to

compare them to see whose difference is significant.

4 RESULTS

In the Results section, there are four parts that help to

answer the research questions mentioned before.

First, the overview of the participants’ physiological

response is reported by showing the mean z-score of

the 16 participants. So we could build a general idea

of the participants’ response of the three videos

according to the first part. After that, the results of the

SCR of the participants are described in the second

part, and the results of the SCL are reported in the

third part. Both the general response to different

videos and the responses to specific scenes are

analysed, enabling us to understand the participants’

reaction to both the entire videos and the specific

moments in the videos like the food close-up. And

then, the results of the questionnaires are shown in the

last part of the section, helping us to compare the

results of the objective measurement (GSR sensor)

and the subjective measurement (questionnaire).

4.1 Overview of the Participants’

Physiological Response (R1)

First, the mean Z score levels corresponding to the

time, analysed from the data of 16 participants during

the whole experiment, are shown in Figure 5. An

overview of the participants’ physiological response

to the three videos could be thus built.

In Figure 5, there are three lines representing the

Z score of the participants’ galvanic skin response to

each video. Some descriptions of the moments in the

video corresponding to the peaks are also exhibited in

the figure. The peaks mean that the participants had

the emotional arousals at this moment.

Figure 5: The mean of Z score for each video. The red line

represents the Video A, which promotes the blueberries.

The green line represents the Video B, which addresses the

story of making ham. The blue line shows the Video C,

which advertises the beer chicken. The X-axis is the time

and the Y-axis is the GSR Z score. The red annotations are

corresponded to the peaks, indicating that the participants

had the emotional arousals.

From Figure 5, we can safely conclude that the

participants’ physiological response has three trends:

1) The participants’ response decreased during all

three videos, indicating that the participants were

losing their interests while watching the videos,

especially during the Video A. 2) It is easy to find that

the participants had the emotional arousal to the

close-up of the food when watching all videos. 3)

Besides, the arousal level in the three videos varies.

According to the mean Z score, the arousal level in

the Video A is higher than the Video C, and the

arousal level in the Video B remains the lowest.

4.2 SCR Related Video Events

(R1&R2)

Based on the algorithm mentioned in the software

section, event related significant emotional arousal

were detected as shown in Figure 6, 7, and 8.

The SCR results from the three videos are rather

different. The participants were significantly aroused

in almost the entire watching process of the video C

(Figure 8). As a contrast, only two moments caught

the significant arousal in the video B (Figure 7)

and no significant arousal appeared in the video A

(Figure 6).

In addition, the SCR analysis also reveals which

events significantly aroused the viewers. For

instance, in the Video B, the participants were highly

engaged in the scenes where the giant showed the

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

Figure 6: The extracted SCR signals of the participants

during the Video A. The x-axis is the time in seconds, and

the y-axis is the mean p value of the bilateral MWW test.

The blue line is the participants’ mean p value, the red line

represents the critical value (p = .05). When the blue line is

below the red line, it indicates that the participants had a

significant emotional arousal response at 0.05 p value level.

Figure 7: The extracted SCR signals of the participants

during the Video B. The x-axis is the time in seconds, and

the y-axis is the mean p value of the bilateral MWW test.

The blue line is the participants’ mean p value, the red line

represents the critical value (p = .05). When the blue line is

below the red line, it indicates that the participants had a

significant emotional arousal response at 0.05 p value level.

ham to his friend, and the slogan and logo shown up

at the end of the video, except for the moment (around

10 seconds) that they saw a sleeping cat. For the video

A, the participants did not show any significant

emotional moments. While for the video C, only two

events significantly stimulated the participants’

emotions.

4.3 SCL Levels in Three Videos

(R1&R2)

The user SCL level differences in the videos are

shown in Figure 9 and Figure 10, and the significant

bio-responses were observed in them.

Figure 8: The extracted SCR signals of the participants

during the Video C. The x-axis is the time in seconds, and

the y-axis is the mean p value of the bilateral MWW test.

The blue line is the participants’ mean p value, the red line

represents the critical value (p = .05). When the blue line is

below the red line, it indicates that the participants had a

significant emotional arousal response at 0.05 p value level.

Figure 9: The mean and standard error of the SCL data of

each video.

Figure 10: The mean and standard error of the SCL data of

each scene.

The SCL of the participants in the three videos are

significantly different (F (2, 14) = 393.36, p < .05).

By using the LSD method to do the pairwise

comparison, the participants’ SCL data is found to be

significantly different from each other. It shows that

How Commercial Food Videos Affect Female Customers - Measuring Female Bio-response Towards Commercial Food Videos

the SCL is the highest in the video C, while it is the

lowest in the video B. It is also spotted that the

viewers’ SCL showed significant differences among

the six scenes where the close-up scenes occurred

(three from the blueberry video, and only one close-

up from the other two videos respectively): F (5, 11)

= 53.62, p < .05). The LSD method shows that the

SCL data is significantly different from each other,

except for the second close-up of the blueberry and

the one of the beer chicken. In particular, the SCL of

the participants during the first close-up of the

blueberry is significant higher than the other scenes.

4.4 Questionnaires (R3)

There is no significant difference among the

evaluation of the videos: F (2, 14) = 5.20, p = .11

based on the repeated measuring ANOVA test of the

questionnaire.

Figure 11: The disliking ratio reported from the

questionnaires.

Figure 12: The liking ratio reported from the

questionnaires.

The ratio on the likes and dislikes videos shows

in Figure 11 and 12. More than a half of the

participants disliked the video B because of the dark

image and scary background music. For the most

favourite video, 44% of the participants chose the

video C, 37% of the participants chose the video A,

and only 19% of the participants chose the video B.

The results of the questionnaire are similar to the

results of the SCR and SCL results. For example, both

the SCR and SCL results indicate that the video C is

the most engaging one while the video B is the least

engaging one. To sum up, the results obtained from

the sensors are consistent with the ones obtained from

the questionnaires.

5 DISCUSSION

In this paper, we conducted an experiment by using

GSR sensors, where user bio responses towards the

three food TVCs were analyzed. Through a

comparison with the questionnaires, it has been

confirmed that the sensor measurement is consistent

with their subjective reports.

The study has answered the research questions

clearly. First, according to both the GSR results and

the questionnaires, the participants preferred the

video C (Beer Chicken) the most, and showed the

least preference to the video B (Ham). Second, some

specific scenes, like the close-up of the food, could

induce the participants’ emotional arousal. Third, the

results of the GSR data were grounded by the

questionnaires.

There are some interesting findings discovered

from the results of the study. First, the attention in all

videos is decreasing from the start to the end of each

video according to the z-score figure, which means

that the participants were losing their interest during

all the videos, especially after 90 seconds. It suggests

that the length of a TVC is crucial. This information

is extremely helpful for producers, especially when it

comes to the effectiveness of a TVC. Second, some

interesting phenomena were observed among the

female consumers. One of them is that they prefer

some specific moments in a video, e.g., the close-up

of the food, the good-looking figure of the actors, and

the appearance pets. These reasons may explain why

the video C is the most popular one. In addition, it

seems that women are sensitive to the bright scene

and the joyful music background, which were

reported from the post-interviews.

Nevertheless, there are some differences found

between the SCL and the SCR. According to the

results of the SCL, the arousal level of the video C is

the highest, while the video B is the lowest, which is

consistent with the results of the questionnaire.

However, according to the results of the SCR, there is

no significant arousal moments during the video A,

which means the video A did not significantly arouse

the viewers. We assume that the two patterns of the

EDA may reflect the different user experiences, and

we need further investigation for such phenomenon.

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

This study proves again that our GSR sensor

system is robust for such studies. In particular, the

monitoring system has helped us learn the actual

status of sensors. Even if the sensor gets broken, we

can easily replace it with a new one. However, the

sensor prototype is a bit of bulky for users to wear. A

half size of the current one will be an optimal shape

for an unobtrusive watching experience.

In this study, we did not categorize which specific

emotions are induced in viewers, considering the fact

that we only have one type of sensors in our case.

Therefore, more sensors are required if user

emotional states need to be further classified. In the

future work, we can add other types of sensor (e.g.,

ECG sensors) to obtain more sensor data, which may

help us better define emotions elicited during video

consumption.

6 CONCLUSION

In this paper, we have reported the experiment on

female bio-response towards the three types of food

TVCs. The results have exhibited how those videos

could affect female reactions and their watching

experiences. Our study presents that physiological

data does have superior advantages on measuring user

experiences compared to subjective reports. By

following our method, researchers can design an

experiment with their own research purposes.

Furthermore, other similar studies, e.g., new media

design, can be also benefited from our learning

experience.

Besides, our work also demonstrates that the

combination of the hardware and software solution

can be rather helpful for commercial companies. By

using our method, they can pre-assess the effects of

TVCs, especially among targeted consumers who are

particularly interested in investigation. In such a

manner, it can reduce the risk before the launch of

products, and pre-sampling test method can help them

to adjust the marketing strategy.

In addition, it has been fully demonstrated that our

GSR sensor system is robust and can simultaneously

and accurately capture the GSR signals from users.

The system allows us to quantify user experience, and

at the same time keeps the confidentiality with user.

Currently we are working on the process of scaling up

the system and attempting with the integration of

different sensors

(e.g., ECG sensor and acceleration

sensors). The other types of sensor can be integrated

into our sensor network to provide a more complete

representation of user experience.

REFERENCES

DualShock. (2017). [online] En.wikipedia.org. Available

at: http://en.wikipedia.org/wiki/DualShock. Factors in

Computing Systems (CHI '13). ACM, New York, NY,

USA, 1499-1508.

Smith, J., 1998. The book, The publishing company.

London, 2

edition.

Aigner, W., Tomitsch, M., Stroe, M. & Rzepa, R. Be a

judge!: wearable wireless motion sensors for audience

participation. CHI '04 Extended Abstracts on Human

Factors in Computing Systems, 2004. 1617-1621.

Barkhuus, L. Engaging the crowd:studies of audience-

performer interaction. Extended Abstracts Proceedings

of the 2008 Conference on Human Factors in

Computing Systems, CHI 2008, Florence, Italy, April,

2008. 2925-2930.

Chanel, G., Rebetez, C. & Pun, T. Boredom, engagement

and anxiety as indicators for adaptation to difficulty in

games. International Conference on Entertainment and

Media in the Ubiquitous Era, Mindtrek 2008, Tampere,

Finland, October, 2008. 13-17.

Fischer, J. E. & Benford, S. Inferring player engagement in

a pervasive experience. Sigchi Conference on Human

Factors in Computing Systems, 2009. 1903-1906.

Fleureau, Julien, Philippe Guillotel, and Izabela Orlac.

2013. "Affective benchmarking of movies based on the

physiological responses of a real audience." Affective

Computing and Intelligent Interaction (ACII), Humaine

Association Conference on. IEEE.

Hart, J., Sutcliffe, A. & De Angeli, A. Using affect to

evaluate user engagement. CHI '12 Extended Abstracts

on Human Factors in Computing Systems, 2012. 1811-

1834.

Joho, H., Jose, J. M., Valenti, R. & Sebe, N. Exploiting

facial expressions for affective video summarisation.

ACM International Conference on Image and Video

Retrieval, Civr 2009, Santorini Island, Greece, July,

2009. 1-8.

Kapoor, A., Burleson, W. & Picard, R. W. 2007. Automatic

prediction of frustration. International Journal of

Human-Computer Studies, 65, 724-736.

Kenning, P. H. & Plassmann, H. 2008. How neuroscience

can inform consumer research. IEEE Transactions on

Neural Systems & Rehabilitation Engineering A

Publication of the IEEE Engineering in Medicine &

Biology Society, 16, 532.

Krugman, D. M., Fox, R. J., Fletcher, J. E., Fischer, P. M.

& Al, E. 1994. Do adolescents attend to warnings in

cigarette advertising? An eye-tracking approach.

Journal of Advertising Research, 34, 39-52.

Latulipe, C., Carroll, E. A. & Lottridge, D. Love, hate,

arousal and engagement:exploring audience responses

to performing arts. International Conference on Human

Factors in Computing Systems, CHI 2011, Vancouver,

Bc, Canada, May, 2011. 1845-1854.

Lin, T., Maejima, A. & Morishima, S. Using subjective and

physiological measures to evaluate audience-

participating movie experience. Working Conference

How Commercial Food Videos Affect Female Customers - Measuring Female Bio-response Towards Commercial Food Videos

on Advanced Visual Interfaces, AVI 2008, Napoli,

Italy, May, 2008. 49-56.

Lunn, Darren, Harper & Simon 2010. Using galvanic skin

response measures to identify areas of frustration for

older web 2.0 users.

Madan, A., Caneel, R. & Pentland, A. S. GroupMedia:

distributed multi-modal interfaces. International

Conference on Multimodal Interfaces, 2004. 309-316.

Mirza-Babaei, P., Nacke, L. E., Gregory, J., Collins, N. &

Fitzpatrick, G. How does it play better?:exploring user

testing and biometric storyboards in games user

research. Sigchi Conference on Human Factors in

Computing Systems, 2013. 1499-1508.

O'Brien, H. L. & Mclean, K. 2009. Measuring the User

Engagement Process.

Oliveira, E., Benovoy, M., Ribeiro, N. & Chambe, T.

Towards emotional interaction: using movies to

automatically learn users' emotional states. Ifip Tc 13

International Conference on Human-Computer

Interaction, 2011. 152-161.

Pan, M. K. X. J., Chang, J. S., Himmetoglu, G. H., Moon,

A., Hazelton, T. W., Maclean, K. E. & Croft, E. A.

Galvanic skin response-derived bookmarking of an

audio stream. International Conference on Human

Factors in Computing Systems, CHI 2011, Extended

Abstracts Volume, Vancouver, Bc, Canada, May, 2011.

1135-1140.

Peters, C., Castellano, G. & Freitas, S. D. 2009. An

exploration of user engagement in HCI, ACM.

Picard, R. W. & Daily, S. B. 2005. Evaluating affective

interactions: Alternatives to asking what users feel. Chi

Workshop on Evaluating Affective Interfaces

Innovative Approaches.

Sakurazawa, S., Yoshida, N., Munekata, N., Omi, A.,

Takeshima, H., Koto, H., Gentsu, K., Kimura, K.,

Kawamura, K. & Miyamoto, M. A computer game

using galvanic skin response. International Conference

on Entertainment Computing, 2003. 1-3.

Sas, C., Fratczak, T., Rees, M., Gellersen, H., Kalnikaite,

V. & Coman, A. Affectcam: Arousal-augmented

sensecam for richer recall of episodic memories.

CHI '13 Extended Abstracts on Human Factors in

Computing Systems, 2013. 1041-1046.

Wang, H., Prendinger, H. & Igarashi, T. Communicating

emotions in online chat using physiological sensors and

animated text. CHI '04 Extended Abstracts on Human

Factors in Computing Systems, 2004. 1171-1174.

Yoon, K., Bolls, P. & Lang, A. 1998. The Effects of arousal

on liking and believability of commercials. Journal of

Marketing Communications, 4, 101-114.

PhyCS 2017 - 4th International Conference on Physiological Computing Systems