See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/354545267

Big Five Personality Detection Using Deep Convolutional Neural Networks

Preprint · September 2021

DOI: 10.20944/preprints202109.0199.v1

CITATIONS READS

4 2,592

2 authors:

Waiel Tinwala Shristi Rauniyar

Delhi Technological University Delhi Technological University
2 PUBLICATIONS 4 CITATIONS 3 PUBLICATIONS 19 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Waiel Tinwala on 13 September 2021.

The user has requested enhancement of the downloaded file.

 Big Five Personality Detection Using Deep
Convolutional Neural Networks
Waiel Tinwala Shristi Rauniyar
Dept. of Computer Science and Engineering Dept. of Computer Science and Engineering
Delhi Technological University Delhi Technological University
Delhi, India Delhi, India
[email protected] [email protected]

Abstract—Personality is the most critical feature that tells competent, self-disciplined and driven to achieve
us about an individual. It is the collection of the individual’s their goals. If this trait is not dominant, then it tells us
thoughts, opinions, emotions and more. Personality detection is that the individual is careless, procrastinates and is
an emerging field in research and Deep Learning models have disorganized.
only recently started being developed. There is a need for a
larger dataset that is unbiased as the current dataset that is  Agreeableness (AGR): This trait is all about how an
used is in the form of questionnaires that the individuals individual treats others in relationships. Individuals
themselves answer, hence increasing the chance of unconscious dominantly having this trait are perceived as straight-
bias. We have used the famous stream-of-consciousness essays forward, modest, sympathetic, and forgiving. On the
collated by James Pennbaker and Laura King. We have used other hand, individuals having a low measure of this
the Big Five Model often known as the five-factor model or trait are perceived as demanding, insulting, stubborn
OCEAN model. Document-level feature extraction has been and unsympathetic.
performed using Google’s word2vec embeddings and Mairesse
features. The processed data has been fed into a deep
convolutional network and a binary classifier has been used to
classify the presence or absence of the personality trait. Hold-
out method has been used to evaluate the model, and the F1
score has been used as the performance metric.

Keywords—Personality Detection, Big Five model, Five-

factor Model, OCEAN, Deep Convolutional Neural Network,
Mairesse Features, stream-of-consciousness

I. INTRODUCTION
Personality is defined as nature, characteristics or
qualities that form an individual’s character. It is the
combination of the individual’s thought process, opinions,
emotions, fears, goals, and aspirations.
The personality of the individual is the factor that
determines how an individual is going to be with their family
and friends, what kind of career they will opt for, how they
will perform at work and exist in the global community.
When individuals are well-versed with their personality, they
know about their strengths and weaknesses which will help
in self-development.
Some widely used theories for classifying personality
traits are the Big Five model, Myers-Briggs Type Indicator,
Keirsey Temperament Sorter, Three-trait personality model
and Eysenck Personality Questionnaire [1] .
Fig. 1. The Big Five Model
In this paper, we are going to classify personalities on the
Big Five classification. This classification remains relatively  Extraversion (EXT): This trait describes the
uniform throughout one’s lifetime and is the most widely tendency of an individual to be in constant social
accepted personality classification theory adapted by interaction and be interacting with their environment.
psychologists and researchers worldwide. Having a high measure of extraversion means the
individual is very outgoing and enjoys being the
The Big Five model classifies personalities under five
centre of attention. Having a low measure of
broad umbrella terms that encompass more labels. The five
extraversion means that the individual is reserved in
personality categories are described below:
nature, prefers solitude over social interaction and
 Conscientiousness (CON): This trait describes the does not want to be the centre of attention.
individual’s ability to control impulses for goal-
 Openness (OPN): This trait reflects an individual’s
oriented tasks. It describes the ability of the
inclination towards trying new things and engaging in
individual to be organized, responsible, hard-working
new, intellectual and creative activities. An individual
and adhering to the rules and regulations. If this trait
having a high measure of openness means they are
is dominant in an individual, the individual is
creative and think outside of the box. A low measure

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

 of this trait indicates that the individual prefers Five model. The used Support Vector Machines, Random
routine and traditional ways over new methods. forests, Logistic Regression, C4.5 Decision Tree and k-
nearest neighbours algorithm.
 Neuroticism (NEU): This trait tells us about the
overall emotional stability of the individual. It tells us Willy et al. in 2019 [7] have implemented C4.5 Decision
how the individual is likely to interact and react to Tree for classification of personality based on the Big Five
events. Having a high measure of this trait means that Model. They have used the Twitter API to obtain the dataset
the individual is likely anxious, easily irritable and that is roughly around 110 million tweets every day. They
experiences mood shifts. An individual with a low were able to obtain an accuracy of 64.30%.
measure of this trait is calm, confident and resilient.
In 2019, Tao Hong et al. [8] conducted a research to
The rest of this paper is organized as: related works, classify sentiment and detect personality. They have used the
methodology, experimental results, conclusion, future MDSTC dataset (comprises of speech, facial expressions and
work, acknowledgment and references. Galvanic Skin Response) and applied Deep Neural Network
to classify personality. They achieved better results on
II. RELATED WORKS performance metrics as compared to the state-of-the-art
The first Personality Test ever conducted is commonly models.
known as Woodworth Psychoneurotic Inventory and was Imanuel Buhapoda Drexel [9] has used Gaussian Naïve
developed during World War I. The U.S. military used it for Bayes classifier to detect personality on the basis of Big Five
screening army recruits. The screening was done for Post Model. They have applied Word2Vec embeddings as well as
Traumatic Stress Disorder (PTSD). The Process FastText word embedding and used k-fold cross validation
Communication Model (PCM) was widely used. PCM was (k=5) for cross validation.
built by Taibi Kahler and was used for the selection of
astronauts [1] . Songcheng Gao et al. in 2020 [10] have used a Multi-
View Learning Approach that also includes a Multi-Task
The baseline methods of detecting personality are text, approach with StudentLife dataset that they obtains by
audio and visual cues. When using textual data, data developing a mobile application. They collected data from a
preprocessing is a crucial step that directly affects the results. group of 183 individuals from two different universities. The
Generally, the textual features are extracted from the raw data collection is in the form of a questionnaire that one has
data and fed as input into Machine Learning models like to fill out on the mobile application. They have used mean
Support Vector Machines (SVMs), Naïve Bayes Classifier, absolute error and root mean squared error as the
etc. The word embeddings are represented as vectors performance measure.
(Word2Vec, GloVe, etc.) and later more techniques are
applied to yield results. Marwa S. Salem et al. [11] have applied Multinomial
Naïve Bayes, K-Nearest Neighbour, Support Vector Machine
Kasula Chaithanya Pramodh et al. in 2016 [2] used the and Decision Trees on Egyptian Twitter Users Dataset to
stream-of-consciousness and MyPersonality dataset. The classify personality. According to their research, K-Nearest
MyPersonality dataset is a compilation of 250 users updating Neighbor was the best algorithm for Openness, Extraversion,
around 10,000 Facebook Status Updates. They have used Agreeableness and Neuroticism whereas Decision Tree
Natural Language Toolkit for their model and their F1-scores performed best for Conscientiousness.
are 0.665, 0.632, 0.625, 0.624 and 0.637 for OPN, CON,
EXT, AGR and NEU traits respectively. Kamal El - Demerdash et al. [12] have used Universal
Language Model Fine-Tuning for personality trait detection
Carles Venture et al. [3] have used Convolutional Neural on the Big Five model. They have applied this Model on the
Networks in combination with Action Unit Recognition stream-of-consciousness dataset and they have managed to
Systems and face detection to classify personality as per the achieve around 1% better accuracy than most state-of-the-art
Big Five model. They have used the First Impressions models.
dataset that comprises of around 10,000 videos of various
individuals. In 2017, a state-of-the-art research was performed by
Navonil Majumder et al. in 2017 [13] where the researchers
Gokul K et al. [4] in 2018 have applied Bayes-Net used a Deep Convolutional Neural Network to classify
classifier to classify personality as extrovert or introvert. personality traits basis on the Big Five model. The model is
They have used self-recorded audio samples as their dataset. coded in Python 2.7 and uses Google’s word2vec
Auditory Nerve Modeling that comprises of Voice Activity embeddings and Mairesse features. This research was
Detection in combination with Attribute Selection for improved by Md. Abdur Rahman et al. in 2019 [14] , where
dimensionality reduction was applied and they achieved an the researchers applied different activation functions like
accuracy of 88.3%. Tanh, ReLU and leaky ReLU and observed significant
Bojan Simoski et al. [5] have developed a novel improvement on the results. Even this was coded in Python
technique called Social Contagion Model for classifying 2.7.
personality on the Big Five Model. They used the responses We have coded the same model in Python 3.8 since
of 25 individuals to the Big Five Questionnaire as their Python 3.8 since python 2.7 is no longer supported (as of
dataset. January 1, 2020). The methodology is explained in detail in
Abir Abyaa et al. in 2018 [6] have used the StudentLife the following section.
dataset that comprises of data from a group of 48 students.
They have made use of supervised learning algorithms for
the purpose of classification of personality as per the Big
 III. METHODOLOGY
A Deep Convolutional Neural Network (CNN) has been
trained to perform Personality Classification. This approach
classifies five different personality traits: CON, AGR, EXT,
OPN and NEU using the same network architecture. The
methodology is described in detail in the following sections:
A. Dataset
The stream-of-consciousness essay dataset has been used.
It is an assortment of 2,468 essays that are composed by
people in a controlled environment. The writers also marked
Fig. 3. Uniform Distribution Function
their personality traits themselves. The dataset was compiled
by James Pennebaker and Laura King.
2) Second layer—Convolution: Convolutional filters
B. Data Preprocessing along with max pooling were applied to the document in
Since the dataset contains of noise, we have performed order to extract n-gram features from each sentence. The
cleaning by expanding abbreviations, removing unnecessary obtained feature vectors were concatenated into a sentence
symbols and converting the text to lower-case. After vectors.
performing cleaning, sentence-splitting was performed. The 3) Third layer—1-max pooling: The sentence vectors
unique words are complied into a list that we have termed as from Layer 3 were concatenated into document vectors. The
vocabulary. dimension of a sentence vector is the multiple of n-grams.
C. Deep CNN Architecture The document is considered only if it belongs to at least one
feature.
4) Fourth layer—Concatenation: Mairesse features
were concatenated to the document features obtained in
Layer 3 were added to the document vector. Mairesse
features are a set of document-level features such as
Linguistic Inquiry and Word Count Features, utterance-type
features, Medical Research Council features, and prosodic
features.
5) Fifth layer—Activation: Sigmoid, Tanh and ReLU
functions has been used as activation function. Sigmoid
function is explained in (2) and Fig. 4. Tanh function is
explained in (3) and Fig. 5. ReLU function is explained in
(4) and Fig. 6.

  

Fig. 2. Network Architecture [13]

The different layers are shown in Figure 1 and described

below:
1) First layer—Word vectorization using Word2Vecs:
Word-level feature extraction was performed using 
Google’s Word2Vec Word embeddings library and the Fig. 4. Sigmoid Function
unknown words not found in Word2Vec were taken care of
using uniform distribution. Equation (1) and Fig. 2 explain   
the Uniform Distribution Function.
6) Sixth layer—Classification: The sixth and last layer is
a binary classifier that classifies if the given personality trait
   is present or not by using softmax activation function (5)
which is the same as sigmoid function for binary classifiers.

  
 V. COMPARATIVE ANALYSIS
Activation functions are used in neural networks as
outputs for the nodes in the layers. They are used to add non-
linearity to the network. We have used 3 different activation
functions for our model namely: Sigmoid, Tanh, and ReLU.
The state-of-the-art model has been trained for 6 epochs.
The model was evaluated using k-fold cross-validation
(k=10). Due to GPU and CPU constraints, we have trained
our model for 50 epochs and evaluated our model using
hold-out cross-validation technique with 9:1 as the train-test
Fig. 5. Tanh Function ratio.

  

Fig. 8. F1-score comparison of Personality Trait: Extraversion

Fig. 6. ReLU Function

IV. EXPERIMENTAL RESULTS

The constructed dataset is trained and evaluated using
hold-out technique. The performance measure that is used is
F1 score. It is the harmonic mean of Precision and Recall.
True Positive (TP), False Positive (FP), True Negative (TN),
False Negative (FN) are the measures used to calculate
Precision and Recall. F1-score is given by (8):
Fig. 9. F1-score comparison of Personality Trait: Openness
  

  

  

The results that we have obtained are described in Fig. 7.

Fig. 10. F1-score comparison of Personality Trait: Conscientiousness

Fig. 7. F1-scores when applying various activation functions

We can observe that Tanh function performs best for

traits Extraversion, Neuroticism and Agreeableness with F1- Fig. 11. F1-score comparison of Personality Trait: Agreeableness
scores 61.2%, 66.33% and 62.67% respectively. Sigmoid
performs best for Openness and Conscientiousness with F1-
scores of 69.71% and 67.46% respectively.
 Less biased word embeddings could be explored and the
accuracy of the new model could be compared to the current
model.
We have also observed the need of larger and more
accurate datasets. Most ways of collecting data for this field
include handing out questionnaires. Individuals may or may
not answer the questions honestly and there must be a way to
check the credibility of the responses.
ACKNOWLEDGMENT
Fig. 12. F1-score comparison of Personality Trait: Neuroticism We take this opportunity to thank Dr. Ruchika Malhotra
for her valuable guidance throughout this research, from
Sigmoid and Tanh are non-linear activation functions identifying the problem statement to presenting the research.
whereas ReLU is a linear function. We have trained our We would also extend our thanks to Delhi Technological
model using these three activations and compared our results. University for providing us with all the resources that were
The comparison for each of the five traits can be seen in Fig. required to complete this research successfully.
8, 9, 10, 11 and 12. We observe that the non-linear functions
i.e.; Sigmoid and Tanh performed better than the linear REFERENCES
activation ReLU in all cases. We also observe that our model [1] Y. Mehta, N. Majumder, A. Gelbukh, and E. Cambria, “Recent
performs better than the state-of-the-art model. trends in deep learning based personality detection,” Artif. Intell.
Rev., vol. 53, no. 4, pp. 2313–2339, 2020.
VI. CONCLUSION [2] K. C. Pramodh and Y. Vijayalata, “Automatic personality
recognition of authors using big five factor model,” 2016 IEEE
We have been able to successfully code this model in Int. Conf. Adv. Comput. Appl. ICACA 2016, pp. 32–37, 2017.
Python 3.8 and have been able to optimize the previous code [3] C. Ventura, D. Masip, and A. Lapedriza, “Interpreting CNN
structure as well. We have been able to achieve better Models for Apparent Personality Trait Regression,” in IEEE
performance measures than previous state-of-the-art methods Computer Society Conference on Computer Vision and Pattern
Recognition Workshops, 2017, vol. 2017-July, pp. 1705–1713.
as can be observed in Fig. 5. The proposed model had used [4] K. Gokul and S. Lalitha, “Personality Identification Using
k-fold cross-validation technique (k=10) whereas we have Auditory Nerve Modelling of Human Speech,” 2018 Int. Conf.
used hold-out method because of limited resources and yet Adv. Comput. Commun. Informatics, pp. 1731–1737, 2018.
have achieved better results. [5] E. F. M. Araújo, B. Simoski, and M. Klein, “Applying machine
learning algorithms for deriving personality traits in social
TABLE I. F1-SCORES COMPARISON TABLE network,” in Proceedings of the ACM Symposium on Applied
Computing, 2018, pp. 346–349.
Personality Activation Md. Abdur Rehman Proposed [6] A. Abyaa, M. Khalidi Idrissi, and S. Bennani, “Predicting the
Traits Function et al. (2019) method learner’s personality from educational data using supervised
learning,” ACM Int. Conf. Proceeding Ser., pp. 1–7, 2018.
Sigmoid 33.00 60.65 [7] Willy, E. B. Setiawan, and F. N. Nugraha, “Implementation of
EXT ReLU 35.02 59.10 Decision Tree C4.5 for Big Five Personality Predictions with TF-
RF and TF-CHI2 on Social Media Twitter,” 2019 Int. Conf.
Tanh 39.40 61.20 Comput. Control. Informatics its Appl. Emerg. Trends Big Data
Sigmoid 32.02 65.93 Artif. Intell. IC3INA 2019, pp. 114–119, 2019.
[8] T. Hong, X. Sun, F. Tian, and F. Ren, “Sentiment Classification
NEU ReLU 54.50 64.18 And Personality Detection Via Galvanic Skin Response Based on
Tanh 58.14 66.33 Deep Learning Models,” 2019 5th Int. Conf. Big Data Comput.
Commun., pp. 313–317, 2019.
Sigmoid 35.00 61.43 [9] I. B. Drexel, “Feature Engineering and Word Embedding Impacts
AGR ReLU 45.70 62.24 for Automatic Personality Detection on Instant Message,” in
Proceedings of 2019 International Conference on Information
Tanh 46.16 62.67 Management and Technology, ICIMTech 2019, 2019, vol. 1, no.
Sigmoid 32.16 67.46 August, pp. 155–159.
[10] S. Gao, W. Li, L. J. Song, X. Zhang, M. Lin, and S. Lu,
CON ReLU 52.11 62.65 “PersonalitySensing: A Multi-View Multi-Task Learning
Tanh 32.76 63.09 Approach for Personality Detection based on Smartphone
Usage,” pp. 2862–2870, 2020.
Sigmoid 33.20 69.71 [11] M. S. Salem, S. S. Ismail, and M. Aref, “Personality Traits for
OPN ReLU 58.00 66.90 Twitter Users in the Egyptian Dialect Writing Dataset,” Icsie, pp.
206–211, 2019.
Tanh 59.80 66.50 [12] K. El-demerdash, R. A. El-khoribi, M. A. I. Shoman, and S.
Fig. 13. F1 score Comparison Table between state-of-the-art method and Abdou, “Psychological Human Traits Detection based on
our proposed model Universal Language Modeling,” Egypt. Informatics J., no. xxxx,
2020.
[13] T. Yılmaz, A. Ergil, and B. İlgen, “Deep Learning-Based
VII. FUTURE WORK Document Modeling for Personality Detection from Turkish
As observed in this paper, personality detection is a vital Texts,” Adv. Intell. Syst. Comput., vol. 1069, pp. 729–736, 2020.
[14] M. A. Rahman, A. Al Faisal, T. Khanam, M. Amjad, and M. S.
field that requires more research. Application of personality Siddik, “Personality Detection from Text using Convolutional
detection ranges from enhancing voice assistants (Alexa, Neural Network,” 1st Int. Conf. Adv. Sci. Eng. Robot. Technol.
Siri, Google Assistant, Cortana, etc.) to areas like forensics 2019, ICASERT 2019, vol. 2019, no. Icasert, pp. 1–6, 2019.
and health care. It is also used for hiring purposes and
psychological studies.
View publication stats