Graph-based Personality Modeling for Personality-based Recommender System: Do Race and Questionnaire Type Affect Model Testing Result?

Abstract

Personality Elicitation from Text (PET) in personality-based recommender system has weaknesses in that it obliges the users to have social media account and to write status. We solve this problem by proposing novel method that is by means of personality graph. This graph correlates demographic data and personality T-score. In order to know the effect of race and questionnaire differences, we conducted a study by using different questionnaire and race for the model testing. The training data contained multi-races data whereas test data consisted of only Indonesians. Meanwhile to collect the data, training data and test data employed IPIP50 questionnaire and TIPI questionnaire, respectively. The result led us to hypothesize that race and questionnaire do affect model testing result. We also speculate that TIPI and IPIP50 questionnaires are not convertible each other.