Вы находитесь на странице: 1из 6

2019 International Conference on Electrical, Computer and Communication Engineering (ECCE), 7-9 February, 2019

Empirical Study on Personality Trait Classification


by Food Related Preferences
Tasfia Hoque Raqeebir Rab Khushnoor Rafsan Jani Alam
Department of CSE Department of CSE Department of CSE
Ahsanullah University of Science and Ahsanullah University of Science and Ahsanullah University of Science and
Technology Technology Technology
Dhaka, Bangladesh Dhaka, Bangladesh Dhaka, Bangladesh
tasfia9494@gmail.com jishan005@gmail.com tultul23rafsan@gmail.com

Saif Hasan Khan M. A. Wadud Shuvro Umme Zakia


Department of CSE Department of CSE School of Engineering Science
Ahsanullah University of Science and Ahsanullah University of Science and Simon Fraser University
Technology Technology Vacouver, Canada
Dhaka, Bangladesh Dhaka, Bangladesh ummezakia@gmail.com
saif.sk3639@gmail.com wadudshuvro1993@gmail.com

Abstract—Human personality is a combination of the Meat eater persons are more open, have less extent feelings
behavior, emotion, motivation and thinking pattern and has etc. YG Kim et al [2] conducted a pilot test on 335 visitors
great impact on a person's life, health, and other related and studied the personality traits in tourism and hospitality.
preferences. Food preferences provide rich information for They built a measurement model based on relationship
studying personality of a person. In this paper, we conduct an among food neophobia, food involvement, satisfaction, and
empirical study to predict human personality based on loyalty and indicated 16 personalities based on these
restaurant review on food and other related preferences. We measurements. A person who has food neophobia is more
choose the category to ‘judge/perceive’ from the 4 categories of likely has negative effect on satisfaction and loyalty. On the
16 personality traits. A data set is built from a survey of 100
other hand a person with good eater has positive effect on
people based on a questionnaire about their food related
behavior along with standard personality traits. A
satisfaction, and loyalty. They built a measurement model for
classification algorithm is proposed to classify the participant’s food-related personality traits.
personality from his/her food preference and surrounding These researches show the relationship between food and
environment on a restaurant, using reinforcement learning human psychology but there has not been anything that
that utilizes temporal difference, model based, and on policy directly tells us which personality trait a person belongs to by
techniques. We compare our proposed classification results their food choices and food related preferences. So it’s a big
with standard classification solutions for personality detection
challenge for us to tell whether a person is a ‘judge’ or a
to determine the performance accuracy of our proposed model.
‘perceive’? We found that spicy food lovers are more into
Keywords—personality trait, judge, perceive surrounding adventures and thrills which shows whether these people are
environment, restaurant data, machine learning, reinforcement ‘judge’ or ‘perceive’ in nature. So, the objective of this study
learning is to quantify the relationship between a person’s personality
and food habit using classification algorithm of Machine
Learning [ML]. In this paper we made our own food related
I. INTRODUCTION questions from [1] [11] research works and standard
Persons personality starts developing during childhood. It personality test question that are not related to food
can evolve and develop. Personality is important in many preference [10] and classify the personality using
aspects of our life. To understand a person, personality is all Reinforcement Learning [RL]. We will assume the result
we need. A new approach to identify personality trait is from the standard question is true and our food related
based on the food choices or preferences. The psychological questions will be tested by the result of the standard
factors can be used to classify people into certain type of the questions
16 personalities. A lot of researches, surveys, tests and works
have already been done related to food habits and personality II. BACKGROUND STUDY
traits. The research and survey [1] showed that motivation
and intention of organic food choice in restaurant have A. Personality Traits
influence on personality of a person. Food-related
personality traits were significantly depends on the According to theory of psychological types by Carl G.
motivation of choosing food. Food choice motives explained Jung's [7], people can be characterized by: Extraverted (E)
the reasons of choosing organic food products in restaurants. vs. Introverted (I): preference of human general attitude,
Also research work in [4] found that there is a relation with Sensing (S) vs. Intuition (N): preference of believing external
person personality with liking and in-taking of spicy food world or imagination, Thinking (T) vs. Feeling (F):
and showed that spicy food eaters are thrill seekers, lovers of preference of one of the two functions of judging and
dangers activities. A relationship between foods related Judging (J) vs. Perceiving (P): reflects how a person
personality traits, satisfaction and loyalty among visitors implements the processed information. Judging people is
attending food events are explored in [2]. In [6], the authors more organized and stick to his/her rule in life events. On the
found co-relation between people's diet and behavior. People other hand perceiving means that the person is inclined to
who prefer low fat are more dutiful, ordering, and conscious. improve and explore alternative options.

978-1-5386-9111-3/19/$31.00 ©2019 IEEE


In general, the 16 personality traits are categorized based questions. Questions which get more reward will get more
on four questions. These are [8]: importance and vice versa in the case of punishment. Based
on the answers of these questions our method will tell
“Are you outwardly or inwardly focused?” whether a person is a ‘Judge’(J) or ‘Perceive’ (P).
“How do you prefer to take in information?” In this research work, we employ Temporal Difference
“How do you prefer to make decisions?” (TD) Method of RL, a model-based algorithm with a greedy
technique to estimate the policy of its next action. TD
“How do you prefer to live your outer life?” methods adjust predictions about the future at each state of
Based on the first, second and third questions its way before the final outcome is known and it is better
respectively Extraversion (E) and Introversion (I), Sensing than Monte Carlo technique. TD learns the transition
(S) and Intuition (I), Thinking (T) and Feeling (F) are probability T (s1 | (s0, a)) from the pair of current state s0 and
introduced. Judge (J) and Perceive (P) is based on last action a to the next state s1. If T is successfully learned, the
question. We choose judge (J) and perceive (P) from the 16 agent will know how likely to enter next state given current
personality traits because our background studies showed state and action as input. The drawback of model-based
that J/P is more connected to food rather than other 3 algorithms is it becomes impractical as the state space and
personality traits. The noticeable characteristics are [9]: action space grows. In RL agent can be an on policy agent or
off policy agent. But TD uses on policy learning technique
With Judgers for agent where a agent learn the value of next policy based
Follow a timetable and stick to it. on its current action a derived from the current policy,
whereas off-policy agent learns based on the action a
Take time to prepare. obtained from another policy.
Follow the general agreement and make written
III. DATA COLLECTION METHODS
document on those areas that need more work and
discussion. For our research, we designed questionnaire of 22
questions that reflect personality traits (J /P) and their
Analysis the achievements and decisions which are taken relation to food habit. The survey was conducted in a
so far. restaurant for 100 anonymous participants who voluntarily
Agree short time plan and need of closure. provided their food habit related feedback. The survey was
conducted online and offline based on the suitability to
With Perceivers approach the participants. The population is distributed over
Allow time for things to flow, not necessarily follow time various genders and ages to get a perfect range of answers
table. and variety of personality. Questions 1:11 are food habit
related and question 12:22 are based on standard personality
Prefer to bring new ideas and possibilities. trait patterns.
Encourage time for creativity.
A. Food Related Questions
Support self determination and freedom. Food related questions from questionnaires (Q1 to Q11) are
Accept changes in particular direction that are not co-related and given weight accordingly such as W1 to W11.
necessarily impulsive. The rationality between each question to a personality trait of
‘judge/perceive’ (P/J) are provided below:
B. Data mining & machine learning techniques
Data mining is the process of analysis the interesting, Question Logic
useful pattern of large volume of data and predict the Q1. Do you try to avoid food Never & Sometimes>>P
relationship among the data [12]. This area of computer with fat?
science is a combination of statistical analysis and artificial Often & Mostly>>J
intelligence (such as neural networks, machine learning etc) Q2. Do you try to follow Never & Sometimes
with database management system to analyze the large healthy diet? >>P
collections of digital data, known as data sets.
Often & Mostly>>J
In our approach, we used a machine learning technique to
classify persons’ personality based on their food choices Q3. Are you a fiber consumer? Yes >> J
preferences in the restaurant. Machine learning techniques
No >> P
applied in such cases fall into the class of supervised
learning. But supervised learning is not suitable for our case. Q4. If you are having a bad Yes >>P
We have used Reinforcement learning (RL) technique for day and if you have little
our research work. In RL agent learn from experience in the money, do you go to a food No >>J
form of reward and punishment and adapt the new situations place to lighten up your mood?
[13]. It is an ideal learning technique for data mining
scenarios. The agent is continuously sensing the Q5. One of your friends Take it if most of your
environment, which provides the data and makes decisions suggested to take a particular friends do.>>J
using the feedback from its own action and experiences. Our dish , would you rather
Go with your own
method calculates the reward and punishment and by that, it choice of food>>P
increases or decreases the importance of food-related
Q6. Are you adventurous Yes>>J If the answer to a question is ‘judge’, we put J=1 & P=0,
about trying different unique and if the answer to a question is ‘perceive’, we assign P=-1
food? No >>P and J=0. For example, if there are six answers that say the
person is ‘judge’ and five answers that says the person is
Q7. Do you usually go to new I would prefer the new ‘perceive’, then the result would be +6-5=+1. Since it is a
a place to eat or prefer the one>>P positive number, we conclude that the person is a ‘judge’.
known old one? We added equal weight in the beginning to each of the
I would stick with old
the one>>J eleven questions. As the weight will be trained by the
dataset, the result would be obtained by the summation of the
Q8. Are you a slow eater or Slow Eater>>P multiplier of weight and answer (1 or -1 or 0). Fig. 1 shows
fast eater? the flow chart of the algorithm 1.
Fast Eater>>J
Q9. Which one do you do Try new items>>P
mostly?

Known items which you


had before>>J
Q10. Does your eating style Yes>>J
include eating one food in its
No >>P
entirety before moving on to
the next?
Q11. Do you try to finish your Never & Sometimes
food even if you don’t feel like >>P
having it?
Often & Mostly>>J

B. Standard Personality Trait Questions


The personality trait questions are used in standard
personality test. These questions are collected from online
resource [10].

Question
Q12: You often spend time exploring unrealistic &
impractical yet interesting ideas
Q13: You often rather improvise (without preparation) than
spend time coming with detailed plan
Q14: Keeping your options open is more important than to
do list
Q15: You have no difficulties coming up with a personal
timetable and sticking to it
Q16: You feel very anxious/ nervous in stressful situation
Q17: You rarely do something just out of sheer curiosity
Q18: Being organized is more important to you than being
adaptable
Q19: Your home and work environment are quite tidy
Q20: Your travel plans are thought out Fig. 1. Classification algorithm for personality analysis.

Q21: Your work style is closer to random energy spikes


than to a methodical and organized approach B. Proposed Reinforcement algorithm for weight training
We used Reinforcement algorithm TD for weight training of
Q22: You rarely get carried away by fantasies and idea
food-related questions. We have given balanced importance
to each of the eleven questions. The question that can bring
IV. PROPOSED CLASSIFICATION MODEL out more accurate personality will get more importance. The
importance will be given by multiplying weight to it. Every
A. Proposed Algorithm For Classifying Personality time it contributes more to the correct answer it gets a reward
and if it contributes to a wrong answer it gets a punishment
The proposed Algorithm 1 uses the food related
by the rule of reinforcement. The reward is that its weight
questionnaire to obtain the weights related to the personality will be increased a certain amount (+1 or -1) after being
traits of Judge or Perceive.
multiplied by α, where α is learning rate. This weight will be before and after applying our proposed reinforcement
assigned by the proposed reinforcement algorithm 2. Initially algorithm.
we assigned Wi=1 to each food question and α = 0.3. The
weight can be between of value 0 and 2. We made the result more accurate by applying
reinforcement algorithm. The importance of each question is
Questions that give more accurate answer will get more fixed by the reward and punishment. We notice that by the
importance by assigning more weight to it in terms of giving increase of data the accuracy of the system increases.
reward. For large dataset it will get more accurate. The
flowchart of algorithm 2 is shown in Fig 2.

Fig. 3. Allocated weight process.

Fig. 2. Reinforcement algorithm for personality analysis.

C. Performance Analysis
We multiplied the weight to each answer of every food
question and add all the weight together according to
Algorithm 1. If the answer is positive the result is that the
person is judge and if the answer is negative the result is that
the person is perceive. Fig 3 shows the weight allocation of
answers of the given questions asked to one person to predict
his/her personality and get negative result. So the person is
Pperceived. Finally, we compared these two set of
questionnaires result to see if our questions regarding food
are good enough give the right result and the result came
similar in almost eighty percent of the time.
Fig. 4. Accuracy gain with pervious Data.
After applying the proposed reinforcement algorithm
each question gets the deserved importance. We measured We implemented our algorithms in C++ and for data
the accuracy of our classifier. We also compared our system collection we made online survey in HTML. For data
analysis we used Visual Studio 2017 on Intel core i7 with 8 reinforcement algorithm is less than the model after applying
GB RAM. The execution time running the algorithm was the algorithm. Slowly with the increase of data the accuracy
39.987 seconds. increases.
In figure 5 (a) the green line represents the desired
output. The blue line represents the output without using the D. Some Observations
reinforcement algorithm. The green line in figure 5 (b) a) Question quality: If a question's answer is same
represents the data graph after applying the proposed for all judge and perceive for some critical reason then the
reinforcement algorithm. increasing of weight would be invalid. That is why we chose
the questionnaires carefully.

70%
64%

(a)

(c)

Fig. 5. Performance analysis of the proposed algorithm: (a) before


applying RL (b) after applying RL (c) boxplot.

b) Total zero: We took odd number of


questions(eleven questions) to find the result from majority
questions(6 or above). So in the eleven questions there are
six answers that say the person is judge and five answer that
says the person is perceive then the result would be +6-
5=+1, it is a positive number. So we say the person is judge.
Seeing the result we understand if majority of the answers
was judge or perceive. But when weight is multiplied the
answer might be neither positive nor negative. Instead it
could be a zero. So the person would be 50% perceive and
50% judge.
(b) c) Negative Weight: We cannot keep any negative
We observed from Fig 5 (a) that performance of desired weight as negativity and positivity is deciding whether the
output to classify personality is much better than classifier person is perceive or judge. If we keep negative weight we
performance of data without using RL for all size of data. might get the opposite result. We will stop decreasing the
Later when the system learned from the data using RL and weight value after it gets to zero indicating that the question
started coping with the actual result the accuracy to classify should not contribute to the output at that moment until the
personality is increased as shown in Fig 5 (b). In the weight increases by getting reward.
beginning of the graph 5 (b) the blue and the green line does d) Non-confirmative Solution: New questions cannot
not match that much. But for larger data set performance of
be added, and cannot tell personality from other questions,
desired and learned data is very close. The accuracy becomes
Personality detail cannot be told about a person. We can
70% at 90th data. Fig 4 shows that accuracy to predict
personality increases using RL technique than without using only say about one side of the personallity not the other 3
RL technique for larger data. We observed from Fig 4 and sides of 16 personality traits.
fig 5 (c) that accuracy to predict correct personality is 64%
before applying RL technique and it increases to 70% after V. CONCLUSION
applying RL for 100 data. We can see that at first the In this paper, we propose a model of determining a
accuracy of the model before fixing the weight by participant's personality whether he is in type of Judging or
Perceiving using reinforcement learning. We defined these
two types of personality through some researched and
analyzed questions. The questions are formed keeping in
mind of the food factors that are related to psychological
facts. We have collected data from a survey that we
conducted over 100 people and calculated weight to map the
data value feeding our algorithm. We have obtained 70% of
accuracy in determining ‘judge/perceive’ personality with
the reinforcement learning method.
In future, we want to add more questions in the food-
related questionnaire, more characteristic features of human
and use bigger population of participants for model training,
this will help us increasing accuracy of our classifier.

REFERENCES

[1] L. R. Goldberg and L. A. Strycker, “Personality traits and eating


habits: the assessment of food preferences in a large community
sample,” Personality and Individual Differences, vol. 32, no. 1,
pp. 49 – 65, 2002.
[2] YG Kim, BW Suh, A Eves, “The relationships between food-related
personality traits, satisfaction, and loyalty among visitors attending
food events and festivals”, International Journal of Hospitality
Management, vol.29, no. 2, pp. 216-226, 2010
[3] Agarwal, Reeti & Mehrotra, Ankit, “Personality traits and
repatronage intentions after service failure”, Journal of Consumer
Satisfaction, Dissatisfaction and Complaining Behavior, 2017.
[4] N. K. Byrnes and J. E. Hayes, “Personality factors predict spicy food
liking and intake,” Food Quality and Preference, vol. 28, no. 1, pp.
213–221, 2013.
[5] Dr. Hassan Ghorbani, Seyede Maryam Mousavi, “The study impact
of consumer personality traits on brand personality and brand
loyalty”, International Journal of Academic Research, 2014.
[6] K. Lawson, “What the Foods You Eat Say About Your
Personality,” Broadly, 16-Sep-2016. [Online]. Available:
https://broadly.vice.com/en_us/article/evgnam/what-the-foods-you-
eatsay-about-your-personality. [Accessed: 15-Nov-2018].
[7] C. G. Jung, Psychological types. Abingdon, Oxon: Routledge, 2000
[8] I. B. Myers and P. B. Myers, Gifts differing: understanding
personality type. Mountain View, CA: Consulting Psychologists
Press, 1995
[9] “Judging vs. Perceiving,” changingminds.org. [Online]. Available:
http://changingminds.org/explanations/preferences/judging_perceivin
g.html. [Accessed: 14-Nov-2018].
[10] “Free personality test,” 16Personalities. [Online]. Available:
https://www.16personalities.com/free-personality-test. [Accessed: 15-
Nov-2018].
[11] S. Nijhawan, “14 Eating Habits And What They Reveal About Your
Personality,” indiatimes.com, 11-May-2016. [Online]. Available:
https://www.indiatimes.com/lifestyle/self/did-you-know-your-
eatinghabits-reveal-volumes-about-your-personality-254821.html.
[Accessed: 14-Nov-2018].
[12] C. Clifton, “Data mining,” Encyclopædia Britannica, 26-Sep-2017.
[Online]. Available: https://www.britannica.com/technology/data-
mining. [Accessed: 14-Nov-2018].
[13] “Knowledge-Based Reinforcement Learning for Data Mining.”,
Kudenko Daniel, Grzes Marek, International Workshop on Agents
and Data Mining Interaction, 2009, 21-22

Вам также может понравиться