Study of Approaches to Predict Personality Using Digital Twin

Abstract

With a growing proportion of online activities on social networking sites on different mediums like Facebook, Instagram, Twitter, LinkedIn the requirement for personality prediction associated with this online mediated behavior has also increased significantly. The user generated content on social media can be effectively leveraged to record, analyze and predict personality through different psychological approaches like MBTI, Big Five, and DISC. Predicting personality has displayed an intrinsic influence in multifarious domains like career choice, political influence, brand inclination, customized advertising, improvising learning outcomes, recommender system algorithms and so on. The objective of this paper is to stipulate an overview of different strategies used by researchers to predict personality based on the social media usage and user generated content across prominent social media platforms. It was observed that the personality traits can be accurately inferred from user behavior reflected on social media through attributes like status posted, pictures uploaded, number of friends, groups joined, network density, liked content. As of now, Facebook followed by Twitter are the most prominent social media platforms for conducting the study however, the use other social media platforms like Instagram, LinkedIn are expected to increase exponentially for carrying out personality prediction study.

1. Introduction

A digital twin is an imitation of a physical entity that enables cost effective stimulation. It assists organizations to identity optimization opportunities and improves operational performance. By stating it a replica does not mean that it’s simply an object, to be more specific digital twin is a process which is entwined with today’s leading technologies like AI (Artificial Intelligence), IoT (Internet of Things), XR (extended reality) and Cloud. It follows a bi-directional approach i.e. the flow of information is from both the digital entity and physical entity [1].

Digital twin is broadly classified in three types i.e. DT Prototype (DTP), Digital Twin Instance (DTI), Digital Twin Aggregation (DTA). DTA is simply the prototype of the physical asset which is to be created. It consists of designing, processing and analyzing physical entity. (DTI) is the digital twin of each instance of the manufactured product and is used to run tests under different scenarios. Meanwhile DTA as the name suggests is the aggregation of DTIs to determine product capabilities, prognostics and learning [1]. Azad M. Madni et al. in his study classified digital twins in four levels i.e. Pre-Digital Twin, Digital Twin, Adaptive Digital Twin and Intelligent Digital Twin. These four levels were classified based on the level of sophistication and scope [2].

1.1 Digital twins and AI

Both digital twin and AI are mutually beneficial for each other. AI can help digital twin to make more informed decisions in order to make improvement across various operations for e.g. design process can be accelerated by quickly evaluating many possible design patterns. On the other hand, digital twin generates stimulated data that can be used to train AI models. As of today, most of the advanced reinforcement learning is happening in the gaming sector. Implementing digital twin assists ML test scenarios by creating a virtual environment for e.g. in reinforcement learning, it is feasible to create multiple scenarios or perform multiple tests which would be impossible in physical environment [3]. In terms of industrial benefit, digital twin along with AI can help organizations to improve speed, quality as well as efficiency of working operations. Digital Twins is a proficient element to unlatch advanced forms of Artificial Intelligence.

1.2 Digital twin nowadays

In a survey conducted by Gartner during covid-19 pandemic, 31% of the respondents mentioned that they are leveraging digital twin technology to ensure safety of customers and employees by enabling remote asset monitoring to minimize in-person monitoring [4, 5]. For the same, as illustrated in Figure 1, 47% (i.e. 28% and 19%) of the organizations focus to increase their priority in IoT implementations in spite of the problematic effects of covid-19 [4]. Whereas, 35% of the organizations reduced their IoT investment as an impact of Covid-19. This leads to the fact that majority of the organizations seek to invest more in IoT for cost reduction. With IoT being implemented Garter predicted that the one-third of medium to large size organizations will carry out at least one DT by the end of 2023.

Figure 1.

Covid-19 impact on IoT implementation [4].

Digital twins are changing the way of work across different industries. It encourages cost effective model driven decisions by considering every possible scenario instead of relying on direct implementation of real world stimulations. It allows predicting scenarios, testing before manufacturing, foreseeing maintenance failures and many other benefits such as-

• Fault detection

• Improves performance hence productivity via optimization

• Lowers maintenance cost and reduce down time with the help of predictive maintenance

• Increases efficiency during the process of making decision.

1.3 Implementing digital twin

Digital twin begins with procuring both static and dynamic knowledge. After gathering relevant information, a conceptual model or say, mathematical model is designed considering exchange of attributes between physical environment and virtual model. This model virtually represents the physical processes on which tests are performed with the help of experimental data. It is very important to analyze which data should come in and accumulated in the digital twin. For this purpose, physical attributes of the real world are collected with the help of sensors to measure critical inputs and execution of the desired adjustment takes place with the help of accurators. After specifying the data requirement, communication protocols (TCP/IP, Pub/Sub, ERP) are determined for smooth transmission of data. These connections are affirmed with the help of technologies like cloud computing, network communication and security. Data coming from different sources is collected, processed and then aggregated in the database along with the real time data. Then, the virtual model is created by modeling all the input and output factors. This virtual model is capable of getting real time data inputs from the physical environment. By effectively visualizing, monitoring and analyzing data, it enables stimulations to generate a feasible output leading to an efficient decision making [6, 7] (Figure 2).

Figure 2.

Flow chart of basic steps of DT implementation [5].

2. Application in distinct domains

By offering its efficiency and competitive advantage, this technology would benefit wide range of industries like manufacturing, aerospace, health care, personality [8, 9] prediction, automotive and construction sector [1, 10, 11] (Figure 3).

Figure 3.

Digital twin across distinct sectors.

1. Manufacturing: DT technology is used in manufacturing industries to a great extent. Some of the applications are as follows:

   • Product customization: Ensuring effective product customization as per the need of the customer and its impact on production process.

   • Product development: Minimize development cost of the product by testing feasibility before manufacturing.

   • Avoiding heavy failures by predictive maintenance.

2. Automotive Sector: Automotive industries utilize this technology to produce a feasible prototype of the product. All possible scenarios regarding safety, customer interaction, vehicle maintenance are analyzed to diagnose any issue beforehand by creating a virtual model.

3. Construction Sector:

   • Predictive approach for resource allocation and management.

   • Continuous updation of data gives an added advantage over static 3D models.

   • Optimize designs by providing more visibility and considering operational needs in advance.

   • Reduce the chance of delay in construction with the help of stimulation.

   • Uphold deadlines and budget by improvising communication and logistics in supply chain.

4. Aerospace: In 2002, the concept of digital twins was addressed by John Vickers from NASA and today the aerospace sector is already relying on this technology for safe stimulation environment, mapping and interrogating transport system.

5. Healthcare Sector: digital twin technology can help to ensure whether a particular treatment or medical equipment would be suitable for patient undergoing treatment for example, CardioInsight [7] mapping system creates 3D map of heart’s electrical activity. This diagnostic method also got approval from FDA.. It can contribute a lot to move forward the medical sector by:

   • Stimulating dosage effects

   • Creating models for optimized medical equipment

   • Building efficient care delivery models

   • Creating models for evaluating possible risks during surgery

   • Designing care coordination models

   • Virtual surgery can be performed beforehand

6. Personality traits recognition: Analyzing personality helps to predict an assumed unique way in which an individual responds and makes decision in a particular situation and since services and products are customer-driven, it’s important to place customer requirements at the center of business strategy. This idea has evolved industrial mass production to personalized production by understanding human personality traits.

3. Personality assessment fundamentals

The personality model used and prediction algorithm employed to calculate personality trait results are two of the most vital factors for personality assessment.

3.1 Types of personality model

1. Enneagram: The Enneagram consists of nine comprehensive key types that interact in a unique way.

   After analyzing Enneagram number, Enneagram wing comes into picture which determines how Enneagram number uniquely relates to the personality. Enneagram wing generally emphasizes other aspects of individuality where it is common to exhibit different Enneagram type characteristics [12, 13].

   The nine Enneagram types are-The Reformer; The Helper; The achiever; The individualist; The investigator; The loyalist; The enthusiast; The challenger; The peacemaker.

2. DISC: it is an acronym that represents four personality types i.e. D (Dominance), I (Influence), S (Robustness), C(Caution).

3. MBTI: Myers Briggs Type Indicator determines personality of an individual in four dimensions where each dimension is further divided into two contradictory key factors like- Extraversion (E) – Introversion (I), Sensing (S) – Intuition (N), Thinking (T) – Feeling (F), Judging (J) – Perceiving (P).

   These key factors are embedded together to form 16 different personality types as shown in the table above (Table 1):

4. Big five: also known as OCEAN (openness, conscientiousness, extraversion, agreeableness, and neuroticism), describes human personality [14] by considering five features where these five features are closely interrelated where every individual has a score in these five personality feature. Therefore, this inter-relatability of the features in the Big five personality makes it suitable for multitask learning [12, 15].

Sr. no.	Tag	Personality type
1	ISTJ	The Inspector
2	ISTP	The Crafter
3	ISFJ	The Protector
4	ISFP	The Artist
5	INFJ	The Advocate
6	INFP	The Mediator
7	INTJ	The Architect
8	INTP	The Thinker
9	ESTP	The Persuader
10	ESTJ	The Director
11	ESFP	The Performer
12	ESFJ	The Caregiver
13	ENFP	The Champion
14	ENFJ	The Giver
15	ENTP	The Debater
16	ENTJ	The Commander

Table 1.

MBTI personality types.

4. Approaches to determine personality traits

Different methodologies can be utilized for determining personality traits of an individual. One of the simplest approach is personality questionnaires where various personality inventories are developed in order to create concise psychological scales like- TIPI (Ten Item Personality Inventory), 28-item questionnaire, (BFI) Big Five Inventory which consists of 44 items, NEO PI, NEO PI-R personality inventory (the Revised NEO Personality Inventory) and NEO-FFI-3 (NEO Five Factor Inventory-3) [12, 15]. Apart from these questionnaires, social media usage questionnaire in the form of survey along with other approaches were also utilized by many researchers to determine personality traits.

Linear Discriminate Analysis (LDA), PCA (Principal Component Analysis), SVM (Support vector machines), Multinomial Naive Bayes and AdaBoost are some of the most utilized classification algorithms for determining the text emotion. Linear Discriminate Analysis is a robust model used for classification and dimension reduction tool which can also be leveraged in data pre-processing. LDA assumes that data is distributed normally and each class has identical covariance matrices. As the classes are assumed to be linearly separable, multiple discrimination functions addressing several hyper-planes are made to recognize the classes [16]. AdaBoost is one of the most efficient ensemble methods that endeavor to approximate Bayes classifier by combining distinct weak classifiers after adding up the probabilistic predictions. It ultimately gives one single strong classifier. SAMME.R that is a variety of AdaBoost is used for multiclass classification problems [17, 18].

5. Background and previous studies

A survey of previous work review is carried out based on research papers on distinct approaches like questionnaires, profile information, status updates, groups or communities joined on platforms like Facebook, Twitter and LinkedIn. Schrammel et al. [19] reported the outcomes of web-based survey to examine the relationship between information shared by the individuals on online platforms, their usage patterns and personality attributes. 162 completed questionnaires were gathered from different countries around 44% of the people who participated in the study were graduated. Five factor personality scale questionnaire was used. In order to determine behavioral patterns, community specific questions were also included. The results displayed that individuals with high extraversion scores tends to have a greater number of connections. Meanwhile, agreeableness was not related to the number of connections. It was also observed that individuals investing more time online share more personal information on profile [19]. T. Ryan et al. conducted research on self-selected 1158 Facebook users and 166 Facebook non users where participants were aged between 18 and 44. A bundle of 124 questions comprising Big Five Inventory, SELSA-S and NPI-29 along with Facebook usage survey consisting 18 questions was conducted specifically for Facebook users. The data gathered from both Facebook users and non-users was divided into two groups in order to differentiate which characteristics will more likely belong to Facebook users and non-users [20]. B. Zhong et al. [21] in his study explored the relation between personality characteristics and time spent on internet for different purposes like social media, study and other activities. 436 University students participated in this study by answering survey questions on the total internet usage, multitasking experience, need for cognition, Information and Communication Technology (ICT) innovativeness. The one of the two main factors that can be considered in this study is that an individual might not be much prompted to handle and process information on social networking sites. Another factor is that an individual with high NFC could be more efficient and hence gives less time to social networking sites and vice versa [21].

Mohammad Dalvi-Esfahani et al. utilized personality traits as moderators to evaluate the impact of Perspective Taking (PT) and Empathic Concern (EC) on social media addiction (SMA) [22]. 592 high school students between 15 and 18 yrs. participated in this study. The data gathered was analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM). It was observed that behavior associated SMA was reflected more among high income schools [22]. Pavica Sheldon et al. aimed to determine whether Big five traits, contextual age indicators and fear of missing out are significant indicators of addictions associated with social networking sites. A survey of 337 students consisting of 193 women was conducted. In this study the social activity, interpersonal interaction and life satisfaction were estimated by The Rubin and Rubin life position scale. Big Five Inventory-10 was utilized to measure personality attributes and Bergen Facebook Addiction Scale (BFAS) was used to evaluate social media addiction. As a result, a positive correlation was realized between social activity and Snapchat. It was also observed that FOMO and social media addiction were highly correlated [23]. In a recent study conducted by J. Brailovskaia et al. the consequences of Covid-19 pandemic and its association with the social media usage were examined. A total 550 individuals participated where 76.2% were women. 65.3% of individuals belonged to the student category whereas the percentage of employed individuals was 33.5%. It was discovered that higher the burden of the pandemic, the lower will be the self-control which will ultimately lead to excessive social media usage [24] (Table 2).

Publication	Purpose	Input data	Methodology	Conclusion
T. Ryan [19, 20]	• To identify personality traits associated with Facebook users and non-users.; • To determine whether these traits are connected the way individuals utilize Facebook.	1635 self-chosen Australian Internet users out of which 1324 completed the online questionnaire (1158 FB user and 166 FB non users)	FB usage questionnaire and other including BFI.	Facebook users are less conscientious and socially desolate than non-users but were quite often found to be more extraverted and narcissistic as opposed to the non-users.
B. Zhong et al. [21]	To explore the relation between personality characteristics and social media	436 University students consisting 118 males and 318 females (average age = 20)	Survey questions to determine the relationship between the Big Five personality traits and internet usage.	• There is a negative correlation between Need For Cognition (NFC) and SNS usage.; • People with high NFC also tend to add less people to their SNS accounts.
Mohammad Dalvi-Esfahani[22]	To determine the impact of EC and PT on SMA where personality traits were utilized as moderators.	Questionnaire from 592 high school students where 42.1% were male students.	• PLS-SEM approach was used to analyze data.; • Characteristics of personality were estimated with the help of NEO-Five Factor Inventory which consisted in total of 60 items.; • A series of ANOVA analysis was utilized	• EC and PT can negatively predict SMA.; • Extraversion adversely directed the relationships between EC and SMA and PT and SMA.; • ANOVA unveiled that SMA was reflected irrespective of gender but was reflected more on high income schools.
Pavica Sheldon [23]	To examine significant indicators of social media addictions.	Survey of 337 undergraduate students (141 men and 193 women) was conducted with average age = 23.35 and SD = 8.08	• Pearson product–moment correlations and hierarchical linear regressions were conducted where life-position indicators and Big Five personality characteristics with the social media addiction attributes were correlated first.; • Predictors of social networking sites addictions were analyzed with the help of three hierarchical linear regressions.	Positive correlation was observed between FOMO and social media addiction. Social activity and Snapchat were also found to be positively correlated with each other.
Cecilie Schou Andreassen [25]	To identify association between narcissism, self-esteem and habit-forming usage of social networking sites.	Total 23,532 participants aged between 16 and 88 years with mean age = 38.5 years and SD = 13.3.	Open web-based survey consisting-; • Bergen Social Media Addiction Scale (BSMAS); • Narcissistic Personality Inventory-16; • Rosenberg Self-; Esteem Scale	• It was observed that scores for addictive usage of social platforms were higher among young.; • Self-esteem and addictive use of social networking platforms were found to be positively correlated with narcissism. Whereas, negative correlation was there between self-esteem and social media usage.
J. Brailovskaia[24]	To analyze the burden caused by the pandemic on excessive usage of social networking platforms and mental health.	550 individuals from Germany participated in this research out of which, 76.2% were women. Mean age = 27.08 and SD = 6.74.	Participation survey link was sent to 600 randomly selected people between March – May 2020	• Burden caused by the pandemic was positively correlated with excessive social media usage and anxiety whereas sense of control was negatively correlated.; • Anxiety symptoms were also observed to be correlated positively with addictive social media usage.

Table 2.

Personality Prediction using Questionnaire approach.

Content generated by the user on Facebook was associated with the responses given by them in the questionnaire. The Facebook features used to determine personality traits in this research are- Total status updated, Photos uploaded, Total number of groups joined, Number of Facebook friends, Number of Facebook likes, Number of times the user is tagged in photos by other profiles.

Questionnaires are time consuming and there are high chances of a person not giving accurate answers to avoid being judged. Whereas, the usage-based approach only focuses on the time given by the user rather than the activity type. Hence in order to mitigate these limitations, some other aspects of social media usage were included. Another popular approach utilized to determine personality traits is to study user behavior on social media through certain level of activities like posting status, commenting, following certain groups/ communities.

In order to bridge the gap between social networking sites and personality traits with the help of machine learning, J. Golbeck et al. [26] utilized BFI (Big Five Personality Inventory) and Facebook profile information i.e. “About me” and status update of the individuals. Data was accumulated from 279 individuals meanwhile; linguistic analysis was performed on data gathered from only 167 individuals where at least 10 words in the text fields were present. Golnoosh Farnadi et al. [27] in his study utilized Facebook user generated data from myPersonality app to predict personality traits of an individual. 9917 status updates from 250 Facebook users, frequency as well as time of posting were collected from myPersonality project. One or more personality traits were assigned to the user based on the answers given by them in questionnaire. The four features leveraged in this study are LIWC feature (from status updates and other text), time related feature, social network feature (network size and density) and other features like number of statues per user, number of urls and words occurring more than once [27].

Amichai-Hamburger et al. and Vinitzky et al. in their research suggested a strong connection between user behavior on Facebook and personality for this purpose they targeted participants who tend to use Facebook often. The data was gathered from 237 Israeli university students comprising 136 female and 101 male with mean age 22 years. In order to conduct study in accordance with the five-factor model, students were asked to NEO-PI-R and a self-report measure. Apart from this, user generated content on Facebook was also measured and encoded. Four dimensions of users aimed were: basic information, personal information, contact information & education and work information. It was observed that highly extroverted people tends to share less personal information as they are more confident with their social skills than introverted people also, extroversion was positively correlated to the number of friends instead of groups joined. It was also found that highly neurotic individuals less likely post their pictures as compared to less neurotic people [28].

Michael Tadsse et al. conducted a research to predict the personality traits of Facebook users using Big Five model. myPersonality dataset was used with 250 Facebook users and 9917 Facebook updates. Features selected for analysis, in accordance with the personality attributes were- network size, density, transitivity, betweenness and brokerage. It was found that the extraversion personality trait along with SNA (Social Network Analysis) resulted in best prediction accuracy [29]. Y Bachrach et al. attempted to predict personality traits by analyzing Facebook activities of a user by their actions (posts, groups and likes) and their friends for e.g. tagging and size & density of the friendship network [30].

Content generated by the user on Facebook was associated with the responses given by them in the questionnaire. The Facebook features used to determine personality traits in this research are:

• Total status updated.

• Photos uploaded.

• Total number of groups joined.

• Number of Facebook friends.

• Number of Facebook likes.

• Number of times the user is tagged in photos by other profiles.

Correlations were determined between these features utilized by producing plots known as ‘Clustered Scatter Plots’. By including more Facebook features the accuracy of the model improvised. And a clear picture was observed to determine the correlation between each personality trait and Facebook feature for e.g., agreeableness was found to be positively correlated with groups joined, posts liked and total friends. This study primarily focused on the data of individuals who are active on social media and agreed to use the application for personality prediction. This sort of data might lead to selection bias. Also, the total number of posts liked, groups joined were taken into consideration but not the type of content liked or group joined (Table 3).

Publication	Purpose	Input data	Methodology	Conclusion
J. Golbeck et al. [26]	Predicting user personality through information available on their FB profile.	Performed linguistic analysis on 167 out of 279 individuals. (avg. Age = 31.2 years)	Utilized BFI-45 and FB information “About me” and status updates using Facebook API.	Each of the five personality traits were predicted by training m5sup’Rules and Gaussian Processes ML algorithms.
GolnooshFarnadi et al.[27]	To predict personality traits from Facebook status updates, their time stamps and social network attributes.	9917 status updates from 250 Facebook users, frequency as well as time of posting were collected from myPersonality project.	User generated data on Facebook, answers given by the users in questionnaire,; 81 features extracted using LIWC; SVM, KNN and NB are the three ML techniques that were explored in this study.	ML based approaches generalize across domains.; • Accurate models can be build where no training data is present by utilizing training examples from other sources.
Michael M. Tadesse [29]	To predict personality traits via user generated posts on facebook.	myPersonality dataset consisting of 250 Facebook users and 9917 posts.	• OpenNLP was employed for data pre-processing.; • LIWC and SPLICE were used to analyze content of posts.; • Pearson correlation analysis was performed to inspect significant features for personality traits prediction.	• Neurotic users less likely use words relating to positive emotions.; • Negative correlation was found between conscientious individuals and words with negative emotions.; • The best prediction performance was observed for extraversion personality trait by SNA.
Chris Sumner et al. [31]	To determine whether there is any relationship between Big Five Personality traits and Facebook activity.	Facebook profile information from 537 participants from 15 countries. (mean age = 30 years with 349 female and 174 male participants)	• BFI-44 questionnaire along with 79 Facebook data points consisting age, total number of friends, biography and gender.; • LIWC tool was used to analyze user generated content i.e. post and photo caption/ description.; • Zero-order Spearman’s correlation was used for linguistic analysis.	• People with high extraversion and Conscientious tend to use words with positive emotion. A negative correlation was also observed between the use of swear word and; • Conscientiousness while Neuroticism is positively correlated to it.; • A positive correlation was also observed between number of words used in one sentence and Neuroticism as well as Agreeableness.
Y. Bachrach et al. [30]	Determined the aspects of profile by bifurcating Facebook activities in two categories-; Solely based on user’s activity(posting status, posts liked or groups joined) With respect to actions of user and their connections (size and density of the network, total number of posts tagged)	Records of Facebook profile features of 180,000 users.	• Five Factor Model questionnaire.; • Prediction algorithm used-; Multivariate Linear Regression	• Claimed that the prediction accuracy of the model is increased by adding more Facebook features for analysis.; • All the Facebook features used were positively correlated with Extraversion but negatively correlated with Conscientiousness; • A positive correlation was also found between Openness and Neuroticism with total status updates, likes and groups joined by an individual.

Table 3.

Personality Prediction using content generated on Facebook in association with questionnaire.

5.1 Synthesis of previous work carried on twitter data

In a research conducted by Daniel Ricardo Jaimes Moreno et al., twitter data was utilized to predict personality traits. On twitter users generally post content in form of text i.e., tweets. For training purpose, the author utilized PAN CLEF dataset of 152 users containing 14,166 tweets. The prediction of personality was seen as a classification problem where after creating TFIDF matrix, different dimensionality reduction techniques were applied like PCA (Principal Components Analysis), LDA (Linear Discriminant Analysis) and NMF (Non-negative Matrix Factorization) for extracting latent features. The results displayed that for extraverted traits, TFIDF and PCA performed better whereas, LDA technique gave better results for agreeableness, conscientiousness and stability. Therefore, the best performance was performed by LDA technique [32].

Aditi V.Kunte et al. [33] made hypothesis based on twitter dataset using Twitter API. Data pre-processing was done by conversion to lower case, removing stop-words and special characters. Classification algorithms were then applied to the pre-processed data in order to classify user personality in the class labels of Big Five personality prediction model. It was observed that accuracy, precision recall and F1-score were highest for Multinomial NB as compared to other classification algorithms [34].

Pavan Kumar K. N. and Marina L. Gavrilova [34] aimed to determine MBTI personality traits by taking user generated data in form of latest 50 tweets. For this purpose, the author used a combination of TFIDF, GloVe word embedding technique and SVM classifier. The TFIDF document term matrix and GloVe embeddings of the tweets posted by an individual is utilized to construct decision tree ensembles composed of CART (Classification and Regression tress).

It was observed that the prediction accuracy changed across the MBTI dimensions where S-N (Sensation-Intuition) and E-I (Extraversion-Introversion) dimensions were considerably more reliable. Alexia Katrimpouza et al. [35] used questionnaire and educational activities of students on Twitter in order to determine how learning outcomes are correlated to twitter usage. In total, three studies were conducted where twitter activities of students were analyzed during each study. It was seen that the personality characteristics Openness and Conscientiousness were related with the twitter usage in one of the studies. The tools to implement this study are- SML scale (Social Media Learning), Big Five personality test, TAS (Technology Affinity Survey) and ICTL (Communications Technology Learning). Fabio R. Galloa et al. [36] has made hypothesis that NKB (Network Knowledge Base) model can be associated with personality prediction to develop a hybrid model to predict actions and reactions made by an individual in their social networking feeds. A specimen of NKB and stream of news items for each individual was used to train classifier in order to predict if the user will make a move i.e. a certain action in a certain amount of time. For tuning hyper- parameters the different algorithms utilized were- Logistic Regression, One Class SVM, Random Forests, Decision Trees, Multinomial Naive Bayes, and Complement Naive Bayes. The bigger scope of this study is to reduce the pathogenic feeds by analyzing the information flow on social media [36] (Table 4).

Publication	Purpose	Input data	Methodology	Conclusion
Daniel Ricardo Jaimes Moreno [32]	This study focused on the extraction of features by exploring distinct dimensionality reduction techniques.	Utilized PAN CLEF 2015 data set of 152 different users consisting of 14,166 tweets (approx. 100 tweets per user).	• Matrix was transformed using TFIDF; • For extracting latent feature, different dimensionality reduction techniques were utilized- PCA, LDA and NMF.	• Extroverted tweets carried more words on an average in comparison ofTweets from other characteristics.; • LDA displayed better results for agreeableness, conscientiousness and stability.; • Non-negative Matrix Factorization[37] technique gave inefficient results for each trait as compared to other methods.
Aditi V.Kunte [33]	Whether LDA, AdaBoost or Multinomial NB has higher relevance for personality prediction over twitter dataset.	Twitter data was retrieved using Twitter API consisting 2 columns- index and status.	• Big Five personality prediction model was utilized.; • Focused on real time data using Twitter API.	Multinomial NB displayed highest accuracy, precision, recall and F1-score than AdaBoost and LDA.
Pavan Kumar K. N.[34]	To associate MBTI personality traits in accordance with the latest 50 tweets posted by the user.	• Latest 50 tweets made by the user.; • Twitter MBTI dataset was obtained from the PersonalityCafe forum.	• To extract relevant text features TFIDF and GloVE word embeddings technique was utilized.; • SVM was used for text classification.	The prediction accuracy changed across the MBTI dimensions where S-N (Sensation-Intuition) and E-I (Extraversion-Introversion) dimensions were considerably more reliable.
Alexia Katrimpouza[35]	The educational activities of students were utilized throughout the study in order to obtain an interaction model between social media and participants.	• Inputs were taken from different number of students during each phase of the study.; • First study: 19 students; 19 female (aged between 18 and 28); • Second study: 80 students; (aged between 17 and 47); • Third study: 46 students aged between 18 and 33; 1 man and 45 women	• Twitter activity of the students was analyzed with reference to the total number of tweets posted in each study.; • Big Five personality questionnaire along with TAS, ICTL and SML scale was used.	• It was observed that the students who took part in the weekly tasks were more efficacious in class as compared with the ones who did not take part.; • Openness and Conscientiousness were associated with the twitter usage in one of the studies.; • It was found that the proper consumption of twitter activity amplified the learning experience of students by sharing and searching relevant information.
Fabio R. Galloa [36]	Introduced NKB framework for predicting user actions and reactions on the post.	Tweets and follow network of users.	• Big Five model (from the Personality Insights service offered by IBM Cloud); • Time of the day when prediction is carried out Sentiments present in the post, percentage of positive items in the feed, percentage of negative items	• ML algorithms are efficient to make predictions about how a person responds to the content in the feed.; • The best execution was produced by a variant based on Complement Naive Bayes.; • Personality type proved to be single most impactful feature for displaying significant drops in recall.

Table 4.

Predicting Personality using content posted on Twitter.

Utku Pamuksuz et al. and Joseph T. Yun et al. [38] targeted on three brands i.e. McDonald’s, Harley-Davidson and Tom’s Shoes, to determine personality of the brand’s twitter accounts and the followers associated with it. The main objective of the author was to obtain the connection between human personality and brand’s personality on social networking platforms. Crimson Hexagon was utilized to accumulate brand related user generated dated between 13 July 2009 until 1 October 2015. To attain relation between users and brands, cosine similarity measure was leveraged. The results displayed that the personality type of twitter followers of Harley and Tom’s was closer to the brand personality but it was not the same for McDonald’s and its followers. However, this difference was mitigated on removing neuroticism from the analysis [38].

5.2 Synthesis of previous work carried on multifarious social networking platforms

Niels et al. [39] focused on determining whether job related social platforms can be leveraged to predict personality of a person. Both organization and individual seeking for opportunities can benefit with such sort of assessment as personality affects the job performance. In first study, the respondents were bifurcated into two samples i.e. student (mean age = 23.1, SD = 2.78) and employed individuals (mean age = 36.8 and SD = 10.87) whereas in second study, a homogonous sample of consisting total of 97 participants from one organization took part. A different set of Big Five questionnaire was utilized. To some extent, self-rated extraversion and self-presentation were found to be predicted via LinkedIn profile. However, Big five traits were not evaluated significantly beyond extraversion [39]. Shyron Qianyun Ma et al. [40] conducted a study to examine personality traits of LinkedIn users. A sample of 301 LinkedIn users with average age of 33 years was utilized. The dataset consisted of 43.2% of individuals with 11 or more years of experience and 62.8% post graduated individuals. Snowball technique was used where each LinkedIn individual was asked to approach other five connections to join via mail. The results displayed a close relation between extraversion and agreeableness with LinkedIn use. Meanwhile no significant relation was observed between openness and LinkedIn use. It was also found that individuals with high agreeableness were capable of gaining more social bridging capital [40].

Sumer S. Vaid et al. [41] attempted to determine the linkage between personality traits and frequency of using distinct social networking platforms. Big Five and Dark Triad personality was assessed in association with social network specific usage patterns. For this purpose, relation between different types of social media was analyzed followed by personality traits recognition in accordance of individual’s inclination towards social media usage. The data gathered from exploratory and confirmatory studies were then put together to run a progression of ad-hoc analysis on pooled dataset [41] (Table 5).

Publication	Purpose	Input data	Methodology	Conclusion
Shyron Qianyun Ma [40]	To understand the means in which personality assessment impacts the LinkedIn use and features use.	301 LinkedIn users with average age of 33 years participated in the study.	• Big Five 20 item inventory was utilized.; • Internet Social Capital Scale (ISCS) was used to measure perceived social capital.; • Zero order correlation analysis and five parallel multiple regression analysis were performed to test the hypothesis.	• Extraversion and agreeableness were found to be closely related to LinkedIn use.; • Individuals with agreeableness gained more social bridging capital.; • No significant correlation was found between openness and LinkedIn use or feature use.
Niels [39]	To analyze whether user data generated on job related social platforms is efficient to predict personality traits of an individual.	• In first study, respondents were divided into 2 samples i.e. student sample (age range: 19–35) and working sample (age range: 22–65).; • 97 people from one homogonous organization participated in second study.	Different set of questionnaires based on Big five factor model was utilized in both the studies.	Big five traits could not be analyzed significantly beyond extraversion.
Sumer S. Vaid [41]	To examine multiplatform social media and the personality characteristics linked with the use of different sort of social networking sites.	• The data for exploratory (N: 1803) and confirmatory study (N: 2317 was collected from young university students in the US.; • 10 different types of social media platforms were analyzed.; .	Regression analysis was done on pooled dataset.; • PCA was used to characterize use of various types of social networking platforms.; • Cross sectional approach was followed.	• Extraversion as well as Openness were closely related with the Facebook and messaging platforms usage.; • Young adults higher in Neuroticism used image-content sharing and micro-blogging platforms more frequently.; • Conscientiousness was connected to more noteworthy usage of the YouTube platform.; • Media ecologies of young adults was categorized in 4 use dimensions-; Blogs and Virtual Worlds Social Networking Sites and Messaging Platforms Forums and Media-Sharing Platforms Facebook.

Table 5.

Predicting personality using LinkedIn and multifarious social media platforms.

6. Conclusion

Personality can be predicted on the basis of behavioral characteristics or linkage and content-based data available. In this study we have attempted to explore distinct approaches to determine personality traits with respect to the text posted by the user or their activities on social networking sites. In behavior or say activity-based analysis, an individual is susceptible to display different behaviors at different time intervals i.e. a person can act differently in different situations which will ultimately lead to activation of a specific personality characteristic and will assert expression for it at that specific time. Features like groups joined, total number of friends were also utilized by researchers to accurately determine personality traits. Facebook due to myPersonality application was one of the most prominent platforms used. But in 2018 ban was announced for improper control of data and harvesting personal information of users. After this suspension, Twitter became the most utilized platform to gather data and determine personality traits commonly using Myers Briggs and Big Five approaches.