Personality and Emotions in Decision Making and Recommender Systems Marko Tkalčič1 , Marco de Gemmis2 , and Giovanni Semeraro2 1 Johannes Kepler University, Linz, Austria, marko.tkalcic@jku.at, 2 University of Bari Aldo Moro, Italy marco.degemmis@uniba.it,giovanni.semeraro@uniba.it Abstract. In this paper we survey the work on the usage of personality and emotions in recommender systems. Recommender systems are de- signed to support humans making better decisions. It has been shown that personality and emotions account for the variance in human de- cision making. We present various models and acquisition methods for emotions and personality. Furthermore, we showcase examples of effec- tive exploitation of personality and emotions in RS. We present in more details an example of the usage of emotions as implicit feedback for serendipitous recommendations. Keywords: emotions, personality, decision making, recommender sys- tems 1 Introduction Recommender systems (RS) are being developed for assisting humans in making better decisions. Personality and emotions have been shown to account for indi- vidual differences in human decision making [5,12]. While personality describes enduring personal characteristics, emotions change very rapidly. In this paper we survey how personality and emotions have been used to improve RS. 2 Personality in RS Personality accounts for individual differences among users. Several psycholog- ical models of personality have been developed. Among these, the Five Factor Model (FFM) [17] is the most widely used in RS [29]. The FFM is composed of five basic factors: Openness, Conscientiousness, Extraversion, Agreeableness and Neuroticism. These factors can be acquired explicitly (e.g. through ques- tionnaires [10]) or implicitly (most commonly from social media [9,15]). In RS, personality has been successfully used to solve various problems. Including per- sonality in user-similarity measures has helped alleviate the new-user problem [7,23]. The Openness factor has been proved useful to improve diversity [31]. Personality has also been found to correlate with music preferences [21]. Cross- domain recommendations have been tackled using personality [2]. It was also useful in group modeling for group RS [14,20]. Furthermore, it has also been used to model mood regulation music RS [8]. 3 Emotions in RS Unlike personality, emotions change more rapidly and are harder to model and capture. In RS, emotions are modeled either through the model of basic emotions (e.g. the six basic emotions happiness, anger, fear, sadness, disgust and surprise [6]), the dimensional model (i.e. the valence, arousal and dominance dimensions) or the circumplex model [22]. To acquire a user’s emotion in a specific moment we can use either the intrusive questionnaire approach [1] or implicit methods developed in the affective community [11,28]. Emotions have been used in RS in various ways. The role of emotions in the content consumption chain differs in various stages [27]. Affective labeling has been used to improve recommendations [24,25]. The affective state of a user has been used as a contextual feature [13,32]. It has also been shown that personality relates to which emotions the users perceive in watching films [19]. A conversational RS used affective feedback in the form of the hesitation social signal [30]. 4 Focus: Emotions as Implicit Feedback Generally, in the RS literature emotional feedback is mainly associated with multimedia content and it is collected during or immediately after the item consumption. Spontaneous reactions to proposed items are collected with various aims, one of which is to exploit them as implicit feedback for assessing the user’s satisfaction. We argue that affective states derived from facial expressions could be par- ticularly useful in situations where traditional performance measures are not sufficient to catch the perceived quality of suggestions with respect to the spe- cific aspect being assessed. In particular, in [4] we addressed the research ques- tion: Can emotions observed in facial expressions be considered as a trustworthy implicit feedback for assessing the effectiveness of suggestions produced by RS? The investigation was focused on trying to establish/define a ground truth when evaluating the effectiveness of user-centric intelligent services like RS [3]. We started from the (quite obvious) observation that users do not need per- fect rating predictions, but sensible recommendations. Thus, it is important to take into account factors, other than accuracy, which contribute to the perceived quality of recommendations. For example, serendipity of suggestions refers to the capability of providing the user with surprisingly interesting items she might not have discovered by herself. From this perspective, the effectiveness of recommen- dations depends on both attractiveness and unexpectedness of suggested items. While attractiveness is usually determined in terms of closeness to the user profile, the assessment of unexpectedness of recommendations is not immediate since it involves the evaluation of the emotional response of the user. Thus, the problem of assessing the perceived quality of recommendations can be summarized by the following questions: Can we recognize a sensible rec- ommendation by reading the face of the users exposed to it? Can we read (on the face of the user) the pleasant surprise a sensible recommendation induces? Can we model the degree of serendipity conveyed by sensible recommendations by measuring the emotional response of the user? To this purpose, we designed a study with real users aiming at assessing the actual perception of serendipity of recommendations and their acceptance in terms of the widely adopted metrics of relevance and unexpectedness [18]. To measure the degree of satisfaction related to user experience and gather feedback in a movie recommendation scenario, we used both a questionnaire approach based on two simple binary questions (“Did you know this movie?” for assessing unexpectedness and “Do you like this movie?” for evaluating relevance) and an TM implicit affective labeling method implemented in Noldus’ FaceReader , a tool able to detect basic emotions [6] by analyzing videos that record users’ facial expressions. Sensible recommendations were associated to the positive emotions of happiness and surprise. The results of the experiment show an agreement between the explicit posi- tive feedback acquired by means of the questionnaires and the implicit feedback gathered by means of the detection of happiness and surprise in users’ facial expressions, thus revealing that emotions might help to assess the perception of effectiveness of RS as well as to contribute to the creation of a ground truth for the purpose of RS evaluation. 5 Future work There are many open issues in the domain of personality- and affective-based RS. The lack of datasets is a problem that should be addressed (only a handful of these are currently available [15,16,26]). Furthermore, better implicit methods for the acquisition of personality and emotions should be developed. Personality and emotions play different roles at different stages of the process of selection and consumption of content. It is important to develop models, which use emotions and personality, that account for individual differences in the decision making as well as in the consumption and feedback stages of consumption to close the loop of personality and affective recommendations. References 1. M. Bradley and P. Lang. Measuring emotion: the self-assessment manikin and the semantic differential. Journal of behavior therapy and experimental psychiatry, 25(1):49–59, 1994. 2. I. Cantador, I. Fernández-tobı́as, and A. Bellogı́n. Relating Personality Types with User Preferences in Multiple Entertainment Domains. EMPIRE 1st Workshop on ”Emotions and Personality in Personalized Services”, 10. June 2013, Rome, 2013. 3. L. Chen and P. Pu. A User-Centric Evaluation Framework of Recommender Sys- tems. In B. P. Knijnenburg, L. Schmidt-Thieme, and D. Bollen, editors, Proceedings of the ACM RecSys 2010 Workshop on User-Centric Evaluation of Recommender Systems and Their Interfaces (UCERSTI), volume 612 of CEUR Workshop Pro- ceedings, pages 14–21. CEUR-WS.org, 2010. 4. M. de Gemmis, P. Lops, and G. Semeraro. An investigation on the serendipity problem in recommender systems. Submitted manuscript, 2014. 5. M. Deniz. An Investigation of Decision Making Styles and the Five-Factor Per- sonality Traits with Respect to Attachment Styles. Educational Sciences: Theory and Practice, 11(1):105–114, 2011. 6. P. Ekman. Basic Emotions. In T. Dalglesish and M. J. Power, editors, Handbook of Cognition and Emotion, number 1992, pages 45—-60. John Wiley & Sons, Ltd, Chichester, UK, 1999. 7. M. Elahi, M. Braunhofer, F. Ricci, and M. Tkalcic. Personality-based active learn- ing for collaborative filtering recommender systems. AI*IA 2013: Advances in Artificial Intelligence, pages 360–371, 2013. 8. B. Ferwerda, M. Schedl, and M. Tkalčič. Personality & Emotional States: Under- standing User’s Music Listening Needs to Enhance Recommender Systems. sub- mitted to CHI 2015. 9. J. Golbeck and E. Norris. Personality, movie preferences, and recommendations. In Proceedings of the 2013 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining - ASONAM ’13, pages 1414–1415, New York, New York, USA, 2013. ACM Press. 10. S. D. Gosling, P. J. Rentfrow, and W. B. Swann. A very brief measure of the Big-Five personality domains. Journal of Research in Personality, 37(6):504–528, Dec. 2003. 11. H. Gunes, B. Schuller, M. Pantic, and R. Cowie. Emotion representation, analysis and synthesis in continuous space: A survey. In Face and Gesture 2011, pages 827–834. IEEE, Mar. 2011. 12. D. Kahneman. Thinking, Fast and Slow, volume 1. Farrar, Straus and Giroux, 2011. 13. M. Kaminskas and F. Ricci. Location-Adapted Music Recommendation Using Tags. User Modeling, Adaption and Personalization, pages 183–194, 2011. 14. M. Kompan and M. Bieliková. Social Structure and Personality Enhanced Group Recommendation. UMAP 2014 Extended Proceedings, 2014. 15. M. Kosinski, D. Stillwell, and T. Graepel. Private traits and attributes are pre- dictable from digital records of human behavior. Proceedings of the National Academy of Sciences, pages 2–5, Mar. 2013. 16. A. Košir, A. Odić, M. Kunaver, M. Tkalčič, and J. F. Tasič. Database for contextual personalization. Elektrotehniški vestnik, 78(5):270–274, 2011. 17. R. R. McCrae and O. P. John. An Introduction to the Five-Factor Model and its Applications. Journal of Personality, 60(2):p175 – 215, 1992. 18. T. Murakami, K. Mori, and R. Orihara. Metrics for Evaluating the Serendipity of Recommendation Lists. In K. Satoh, A. Inokuchi, K. Nagao, and T. Kawamura, editors, New Frontiers in Artificial Intelligence, volume 4914 of Lecture Notes in Computer Science, pages 40–46. Springer, 2008. 19. A. Odić, M. Tkalčič, J. F. Tasič, and A. Košir. Personality and Social Context : Impact on Emotion Induction from Movies. UMAP 2013 Extended Proceedings, 2013. 20. J. A. Recio-Garcia, G. Jimenez-Diaz, A. A. Sanchez-Ruiz, and B. Diaz-Agudo. Personality aware recommendations to groups. In Proceedings of the third ACM conference on Recommender systems - RecSys ’09, page 325, New York, New York, USA, 2009. ACM Press. 21. P. J. Rentfrow and S. D. Gosling. The do re mi’s of everyday life: The structure and personality correlates of music preferences. Journal of Personality and Social Psychology, 84(6):1236–1256, 2003. 22. K. R. Scherer. What are emotions? And how can they be measured? Social Science Information, 44(4):695–729, Dec. 2005. 23. M. Tkalcic, M. Kunaver, A. Košir, and J. Tasic. Addressing the new user problem with a personality based user similarity measure. 2nd Workshop on User Models for Motivational Systems: The affective and the rational routes to persuasion (UMMS 2011), 2011. 24. M. Tkalcic, A. Odic, A. Kosir, and J. Tasic. Affective Labeling in a Content-Based Recommender System for Images. IEEE Transactions on Multimedia, 15(2):391– 400, Feb. 2013. 25. M. Tkalčič, U. Burnik, and A. Košir. Using affective parameters in a content-based recommender system for images. User Modeling and User-Adapted Interaction, 20(4):279–311, Sept. 2010. 26. M. Tkalčič, A. Košir, and J. Tasič. The LDOS-PerAff-1 corpus of facial-expression video clips with affective, personality and user-interaction metadata. Journal on Multimodal User Interfaces, 7(1-2):143–155, Aug. 2013. 27. M. Tkalčič, A. Košir, J. Tasič, and M. Kunaver. Affective recommender sys- tems : the role of emotions in recommender systems. Proceedings of the Rec- Sys 2011 Workshop on Human Decision Making in Recommender Systems (Deci- sions@RecSys’11), pages 9–13, 2011. 28. M. Tkalčič, A. Odić, and A. Košir. The impact of weak ground truth and facial expressiveness on affect detection accuracy from time-continuous videos of facial expressions. Information Sciences, 249:13–23, Nov. 2013. 29. A. Vinciarelli and G. Mohammadi. A Survey of Personality Computing. IEEE Transactions on Affective Computing, 3045(c):1–1, 2014. 30. T. Vodlan, M. Tkalčič, and A. Košir. The impact of hesitation, a social signal, on a user’s quality of experience in multimedia content retrieval. Multimedia Tools and Applications, Mar. 2014. 31. W. Wu, L. Chen, and L. He. Using personality to adjust diversity in recommender systems. Proceedings of the 24th ACM Conference on Hypertext and Social Media - HT ’13, (May):225–229, 2013. 32. Y. Zheng, R. Burke, and B. Mobasher. The Role of Emotions in Context-aware Recommendation. Proceedings of the RecSys 2013 Workshop on Human Decision Making in Recommender Systems (Decisions@RecSys’13), 2013.