A computational model of emotion and personality in e-learning environments Somayeh Fatahi School of Electrical & Computer Engineering, University of Tehran, Tehran, IRAN Dalhousie University, Halifax, Canada s.fatahi@ut.ac.ir ABSTRACT* model relating the desirability of events to the personality of a user One of the currently most important discussions in artificial in E-learning environments. intelligence is modeling personality and emotion in artificial intelligence, chiefly in Human-Computer-Interaction (HCI). The 2. RELATED WORKS purpose of this research is designing a general model that identifies Several studies have been carried out in order to consider human a user’s affective status based on user’s personality and emotion. characteristics in human computer interaction [7] [8] [9] [10] [11] The proposed model is composed of two main modules: personality [12] [13]. and emotion modules. The personality module detects personality Jin Du and his colleagues [14] handled the model of learner type of a user based on two approaches: determining personality based on the Cattell’s 16 Personality Factor. In 2011, Gong and through users’ actions in a system and using sequential behavioral Wang [9] used Support Vector Machine (SVM) in order to pattern mining to determine personality. In the emotion module, we determine learning styles in the E-learning environment. Haron propose a computational model to calculate a user’s desirability suggested a learning system, including a learning module which based on personality in e-learning environments. The desirability can be adapted to each learner and utilizes fuzzy logic and MBTI of an event is one of the most important factors in determining a personality test [12]. Abrahamian and his colleagues designed an user’s emotions. The proposed model has been evaluated in interface for computer learners according to learners’ personality simulated and real e-learning environments. The results show that type using MBTI test [15]. In [5], a Bayesian network was used to the model formulates the relationship between personality and detect the learners’ learning style in a tutoring system. In [16], the emotions with adequate accuracy. authors presented a framework for automatic detection of the learner's learning style based on the Felder-Silverman model. Keywords: Personality, Emotion, Desirability, Learning Styles Fatahi and her colleagues [17] [18] [19] [20] designed and implemented a virtual tutor and virtual classmate agent that had 1. INTRODUCTION personality and emotion characteristics as a human being. They Nowadays, one of the most used computer applications is in E- used the Ortony, Clore and Collins (OCC) model for emotion learning. The main idea of E-learning is learning anywhere and modeling and MBTI for personality modeling. anytime, but this type of education has brought new problems. It As mentioned above, several studies have been carried out in usually lacks dynamism and does not establish necessary order to consider human characteristics in human computer interactions to attract learners’ attention. interaction especially in E-learning environments. Despite all these For instance, it’s clear that during a learner’s interaction with a efforts, to the best of our knowledge, there is no work modeling the computer, the learner’s emotional states changes [1] which depends relationship between personality and emotion to improve the e- on his/her individual characteristics. Positive emotions play an learning experience. Consequently, we have proposed a model to important role in creativity and flexibility for solving problems show a relationship between personality and one of the most while negative emotions block the thinking process and prevent important variables in determining emotions [6] called desirability. sound reasoning [2] [3] [4]. Then, the learner’s emotional states in Based on the OCC model, a person could alternatively have three E-learning environments must be taken into account [2]. types of focus which are consequence of events, actions of people, Besides, individuals have different personalities, and individuals and aspects of objects [6]. The first type of emotions includes with different personalities show different emotions in facing emotions which are consequences of the events that have occurred. events. Also, individuals with different personalities have different These consequences are obtained according to the desirability or learning style. undesirability level of the events and the person's goals. Based on Consequently, developers ought to concentrate designing the desirability level, the first branch of emotions in OCC model interactive user interfaces based on user’s affective status, and can calculate. personality aiming to make them more realistic and attractive [5]. In addition, in this research, we used MBTI for personality In this paper, we present a computational model to determine the modeling and the OCC model for emotion modeling. The results desirability [6] of events, as one of the important emotions in show the effectiveness of the proposed approaches. human-human and human-computer interaction, based on personality of users. To the best of our knowledge, it is the first * The student was supervised by Hadi Moradi, School of Electrical & Computer Engineering, University of Teheran, Iran, ISRI, SKKU, South Korea 3. PROPOSED MODEL based on the characteristics of an ENTJ type person, the importance In this research, we focus on designing a computational model that value of “develop new skills” and “learn new things” are high and identifies a user’s affective status based on personality and the impact value of “improving their level of competence” is low. emotion. During the user’s interaction with a computer, and The output of Goal-Personality chart to the event appraiser module depending on events happening in the environment and his is set of goals corresponding to the user’s personality. On the other personality, the user’s emotion changes. In this situation, an hand, the output of this module to the desirability calculator is the intelligent system should be responsive to the user’s emotions. The importance value of each goal in determining the desirability based proposed model is composed of two main modules: the personality on the user’s personality. and emotion modules. Event appraiser: The aim of this module is to calculate the Personality module: The goal of this module is detecting impact of environmental events on achieving the goals which are personality type of users based on two approaches: determining determined by the Goal-Personality chart. In other words, the personality through users’ actions in a system and using sequential output of this module determines how much the goals are achieved behavioral pattern mining to determine personality. These based on the events. approaches are automatic. Also, we use MBTI questionnaire to This is based on lot of evidences that show there is a detect users’ personality manually. It should be mentioned that the relationship between personality dimensions and the events that manual method of personality detection is used as a ground truth causes individuals with different personality react in different for automatically personality detection. ways. For example, asking help from a teacher in an e-learning Emotion module: A lot of models have been designed for emotion environment is an environmental event which happened through modeling. One of the most famous ones is OCC model [6] that is learner. This event is related to extroversion/introversion used in most studies. The OCC model calculates intensity of dimension. As we know, extroverted individuals like to ask for help emotion based on a set of variables. The variables are divided into from others and they like to help others. In contrast, introverted two groups: global and local. The desirability is one of the most individuals prefer to perform their tasks alone without help. That is important factors in determining a user’s emotion [6]. why we use the relationship between personality dimensions and In this research, we focus on designing a computational model events to calculate impact of environmental events on achieving the which calculates a user’s desirability based on the user’s goals. personality and his/her goals. The overall architecture of the model Desirability calculator: As mentioned earlier, based on the is illustrated in Fig. 1. In the following, each module is explained OCC model, to calculate the desirability or undesirability level of in more detail. an event, it is essential to know how much an event is in line with a user’s goals. That is why the inputs to this module are the importance values of the goals, provided by the goal-personality chart, and the impact of environmental events on achieving the goals, given by the event appraiser. These inputs are used to calculate the desirability using Eq. 1. ∑𝑛 𝑖=1 𝑙𝑖 ∗ 𝑤𝑖 𝐷𝑒𝑠𝑖𝑟𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = ∑𝑛 𝑙𝑖 ∈ [−1,1] (1) 𝑖=1 𝑤𝑖 in which n is the number of goals, 𝑤𝑖 , given by the Goal- Personality chart, is the importance of the ith goal. Furthermore, 𝑙𝑖 is the achieved level of the ith goal, which is calculated based on the impact of the triggered events and the corresponding personality dimensions. Finally, 𝐷𝑒𝑠𝑖𝑟𝑎𝑏𝑖𝑙𝑖𝑡𝑦 represents how much the triggered events are desirable based on the user’s goals and personality. The desirability is normalized between -1 and 1 in which desirable events range between 0 and 1 while undesirable events range between 0 and -1. 𝑙𝑖 is calculated using Eq. 2 in which Figure 1. The general view of the proposed model the impact of events and the personality dimensions’ levels (pdl) Goals: In every environment, individuals have different goals are considered through a linear relation between these two. which can be determined based on different theories. For example, according to the Ames (1990) theory, in learning environments, 𝑙𝑖 = ∑𝑚 𝑖=1 𝑒𝑖 ∗ 𝑝𝑑𝑙𝑖 (2) there are two categories of people based on their goals: performance motivational orientation (PMO) and mastery in which m is the number of events, and 𝑒𝑖 is the ith event. motivational orientation (MMO). For example, PMO learners have three goals which are “please the teacher and parents”, “do better 4. IMPLEMENTATION AND RESULTS than other colleagues”, and “show a high-level of competence”. 4.1. Personality module Goal-Personality Chart: Based on the fact that there is a We use an E-learning environment in order to evaluate our relationship between goals and personality [21], the most related proposed model. Since Furnham and Jackson clearly expressed that goals and their importance to a personality type can be determined the learner’s learning style is a subset of his/her personality [23] based on expert knowledge. For example, ENTJ type people fall and MBTI is the personality model which has a learning style into MMO category since they always tend to collect information model, we use MBTI learning style model in e-learning and increase their skill levels [22]. Thus, ENTJ type people have environment instead of personality. the three MMO’s goals, i.e. “develop new skills”, “learn new To evaluate proposed personality module, we use a blended things”, and “improve their level of competence”. On the other learning environment for the “Introduction to computing systems hand, each goal, in the related set of goals to a personality type, has and programming” (ICSP) course. The course is taught to the first- its importance value in determining desirability. For example, year students at the school of electrical and computer engineering at the University of Tehran in Iran. The course runs for 18 weeks, agents had different personality types, different goals based on their and 355,155 interaction records were collected of the two hundred personality types, and different levels of knowledge generated and twenty-six students from the Moodle’s log file. Each record randomly based on a normal random distribution. The five events includes "time", "IP address", "action", "URL", "info", "username", generated randomly in the simulated e-learning environment and "first name", "last name" and "email of the corresponding student". the agents responded to them. These event are “finish the proposed The "time" shows the duration of time students did an activity and activities”, “receive appropriate help”, “lack of request for help”, the "info" includes an id uniquely assigned to the page and “low/high effort”. The simulated e-learning environment accessed/used by the students. calculated the desirability of the agents based on Eq. 1 for agents. We define two groups of features: Learning Activity Feature On the other hand, the level of desirability for each agent, based on (LAF) and Context-based Learning Activity Feature (CLAF). To the agent’s activities, personality type, and events, was labeled by determine the best features for predicting the learning styles, we run seven experts as our ground truth. After that, a dataset was collected many important clustering methods through Weka tools. K-means which included: agent’s personality type, agent’s goals, the event (k=2) was the suitable method for separating two MBTI happened, the level of desirability which calculated based on the dimensions. K-means gives best result when data set are distinct or proposed model, and the level of desirability which was labeled by well separated from each other. The results show that there are nine the experts. The dataset was used to train the two FCMs (Fuzzy CLAF of 112 features that separate people with different learning Cognitive Maps) have been designed to modeling relationship styles. Table 1 shows an example of the results. between environmental events, user’s goals, personality Table 1. Context-based learning activity features to dimensions and desirability. These FCMs show the relationship determine LSs in T/F dimension between events, a user’s goals, the user’s personality, and the user’s desirability. After training mode, we test the model to predict Feature Measure Thinking Feeling desirability. To evaluate the model for the PMO and MMO FCMs, we computed 10-fold cross validation on the data set. Results are The Precision 0.61 0.87 reported in table 2. number of Recall 0.94 0.39 Table 2. Accuracy rate of predication of desirability with messages simulated data sent in the Accuracy 0.67 0.67 Personality types The accuracy of the predicted desirability chat rooms F-measure 0.74 0.54 ESFJ 84.26% The number of messages sent in the chat rooms is the best ENFP 85.82% feature to separate feeling students from thinking ones. This feature ESTP 85.57% confirms that feeling people tend to interact and relate to other ENFJ 82.06% students through discussion rooms. Table 1 shows that feeling ISFP 83.27% students used chat rooms more than thinking people. ENTJ 84.69% In the next step, to extract frequent behavioral sequences, we ENTP 81.38% have collected the data from two hundred and fifteen students who ESFP 83.82% registered in the “Introduction to computing systems and ESTJ 82.47% programming” (ICSP) course. We run Generalized Sequential INFJ 80.81% Pattern Mining (GSP) algorithm which is an apriori-based INFP 86.03% algorithm on data. Finally, we found a lot of frequent behavioral INTJ 81.04% sequences in each dimension of MBTI which some of them can be INTP 84.19% meaningful and usable to discriminate individuals. Some examples ISFJ 81.92% of frequent sequences are presented here: ISTJ 83.42% {Results of Quizzes} {Lessons} {Extra Exercises} {Add/Delete ISTP 81.51% Posts}  Introvert In the next step of this research, we tested the model using real {Extra Exercise} {Quiz}{Results of Quizzes} {Lessons}  data. A real e-learning environment is implemented for the Extrovert “Introduction to computing systems and programming” (ICSP) {Review}{Quiz}{Results of Quizzes} {Lessons} Thinking course. We have collected the data from a hundred and thirty-one {Quiz}{Quiz}{Results of Quizzes} {Lessons}{Add/Delete students who participated in this study. Since the number of real Posts} Feeling data which was collected for INTJ, INTP, ISTJ and ISTP type was {Review}{Quiz}{Quiz}{Results of Quizzes} {Extra Exercises} reasonable to test the model, we evaluated the model on these types. Perceiving The results in table 3 show the desirability prediction accuracy for {Lessons}{Quiz}{Add/Delete Posts} {Extra Exercises}Judging this test. {Add/Delete Posts} {Quiz}{Quiz}{Results of Quizzes}iNtuition Table 3. Accuracy rate of predication of desirability with real {Quiz}{Results of Quizzes} {Lessons}{Quizzes}Sensing data The sample of results show that there are different sequences Personality The accuracy of the weights of behaviors for different dimension of learning style. Also, we can INTJ 75.22% predict the learning style of learners based on these sequences with INTP 68.82% high accuracy. ISTJ 75.83% ISTP 68.04% The results in table 2 and 3 show that our hypothesis about the 4.2. Emotion module relationship between events and achieving goals and mapping To evaluate the emotion module, we used a simulated and a between the personality dimensions and achieving goals are correct real e-learning environment. In the simulated e-learning and our expectation in desirability prediction is satisfied. environment, 1878 artificial intelligence agents were used. The 5. CONCLUSION [11] K.-H. Lee, Y. Choi and D. J. Stoni, "Evolutionary algorithm In this research, we focused on designing a user computational for a genetic robot's personality based on the Myers-Briggs model that identifies the user status based on personality and Type Indicator," Journal of Robotics and Autonomous emotions. To evaluate the proposed module, we used a simulated Systems, vol. 60, no. 7, pp. 941-961, 2012. and a real e-learning environment. [12] H. Orozco, F. Ramos, M. Ramos and D. Thalmann, "An In the future, we want to incorporate mood in our modeling to action selection process to simulate the human behavior in improve the desirability prediction by incorporating the virtual humans with real personality," journal of The Visual relationship between mood, emotion, and personality. Furthermore, Computer, vol. 27, no. 4, pp. 275-285, 2011. it is necessary to collect further data to improve the system’s accuracy. Also, modeling of other factors in emotions needs to be [13] R. Santos, G. Marreiros, C. Ramos, J. Neves and J. Bulas- investigated. Cruz, "Personality, Emotion, and Mood in Agent-Based Group Decision Making," Journal of Intelligent system, vol. Acknowledgement 26, no. 6, pp. 58-66, 2011. This work has been partially funded by the Iranian Cognitive Sciences and Technologies Council, grant number 384. [14] J. Du, Q. Zheng, H. Li and W. Yuan, "The research of mining association rules between personality and behavior of learner 6. REFERENCES under web-based learning environment," in In International Conference on Web-Based Learning, 2005. [1] P. Chalfoun, S. Chaffar and F. Claude , "Predicting the [15] E. Abrahamian, J. Weinberg, M. Grady and C. M. Stanton, Emotional Reaction of the Learner with a Machine Learning "The effect of personality-aware computer-human interfaces Technique, Workshop on Motivaional and Affective Issues on learning," Journal of universal computer science, vol. 10, in ITS," in International Conference on Intelligent Tutoring no. 1, pp. 27-37, 2004. Systems, Taiwan, 2006. [16] S. Graf and T. Liu, "Analysis of Learners' Navigational [2] S. Chaffar and C. Frasson, "Using an Emotional Intelligent Behaviour and their Learning Styles in an Online Course," Agent to Improve the Learner’s Performance," in 7th Journal of Computer Assisted Learning, vol. 26, no. 2, pp. International Conference on Intelligent Tutoring System, 116-131, 2009. Brazil, 2004. [17] S. Fatahi and N. Ghasem-Aghaee, "Design and [3] S. A. Jessee, . P. N. O’Neill and R. O. Dosch, "Matching Implementation of an Intelligent Educational Model Based Student Personality Types and Learning Preferences to on Personality and Learner’s Emotion," International Teaching Methodologies," Journal of Dental Education, vol. Journal of Computer Science and Information Security, vol. 70, no. 6, pp. 644-651, 2006. 7, no. 3, pp. 1-12, 2010. [4] B. Kort and R. Reilly, "Analytical models of emotions, [18] S. Fatahi and N. Ghasem-Aghaee, "An Effective Intelligent learning and relationships: towards an affect-sensitive Educational Model Using Agent with Personality and cognitive machine," in In Conference on virtual worlds and Emotional Filters," in In Proceedings of the World Congress simulation, Texas, USA, 2002. on Engineering, UK, 2010. [5] P. García, A. Amandi and S. Schiaffino, "Evaluating [19] S. Fatahi, N. Ghasem-Aghaee and M. Kazemifard, "Design Bayesian networks' precision for detecting students' learning an Expert System for Virtual Classmate Agent," in In styles," Journal of Computers & Education, Elsevier, vol. 49, Proceeding of the World Congress on Engineering, London, no. 3, p. 794–808, 2007. U.K, 2008. [6] A. Ortony, The Cognitive Structure of Emotions, Cambridge, [20] S. Fatahi, M. Kazemifard and G.-A. Nasser , "Design and UK: Cambridge University Press, 1988. implementation of an e-Learning model by considering learner's personality and emotions," In Advances in electrical [7] L. Chittaro and M. Serra, "Behavioral programming of engineering and computational science, vol. 39, no. 1, pp. autonomous characters based on probabilistic automata and 423-434, 2009. personality," Computer animation and virtual worlds, vol. 15, no. 3-4, pp. 319-326, 2004. [21] . Z. Reisz, M. J. Boudreaux and J. O. Daniel , "Personality traits and the prediction of personal goals," Personality and [8] A. Egges , S. Kshirsagar and N. Magnen, "Generic Individual Differences, vol. 55, no. 6, pp. 699-704, 2013. personality and emotion simulation for conversational agents," Computer animation and virtual worlds , vol. 15, no. [22] I. B. Myers and M. H. McCaulley, Myers-Briggs Type 1, pp. 1-13, 2004. Indicator: MBTI, Consulting Psychologists Press, 1998. [9] S. Kshirsagar and . N. Magnenat-Thalmann, "A multilayer [23] A. Furnham, C. J. Jackson, and . T. Miller, "Personality, personality model," in Proceedings of the 2nd international learning style and work performance," Personality and symposium on Smart graphics, Hawthorne, New York, 2002. Individual Differences, vol. 27, no. 6, pp. 1113-1122, 1999. [10] S. Karimi and . M. R. Kangavari, "A Computational Model of Personality," in 4th International Conference of Cognitive Science (ICCS 2011), Procedia - Social and Behavioral Sciences, Tehran, Iran, 2011.