Predicting Personality Traits By Leveraging Social Media

Abstract

It is evident that humans’ behavioral patterns on online social environments reveal important insights into their personalities. Attracted by this fact, the global research fraternity has contributed to this nascent research area in various ways to automatically recognize the personality by observing the digital footprint. But when examining the literature, it reveals that much of the relevant researches are limited to a narrow bound as their ultimate goal is to exploit word level analysis of the text to infer individual personalities. The main drawback of these word level inspections is that they are restricted to explicitly expressed personality factors in the text. However, these social media data contains much more semantically rich information that can provide more insightful information to personality prediction algorithms. A human reading these posts would unravel these hidden relationships with the knowledge and experience they have gained since birth. With the advent of semantic web and related technologies, such knowledge is readily available in the form of taxonomies, ontologies and knowledge bases. Thus, in this study, we intend to exploit the knowledge encoded in structured knowledge bases to derive features from the social media data that are conveyed in a subtle manner in order to provide enriched feature set to the personality prediction algorithms. In other words, the present research work attempts to simulate a portion of the cognitive ability of the human brain by automating the process of identifying important implicit personality factors in a given user generated content. To achieve this goal, enormous amount of knowledge stored in large semantic knowledge bases were manipulated to efficiently predict the personality traits of the users. Concept extraction of the study is achieved by following two major steps. First a semantically augmented text is generated by identifying suitable entities in the user generated content. Secondly, these identified entities are explored in the knowledgebase to derive the suitable concepts related to that entity. Afterwards, these derived concepts are manipulated to obtain the novel knowledge-based features. To evaluate the integrity of these knowledge-based features numerous experiments were conducted by changing the learning algorithms along with their parameter values, nature of feature values and attribute evaluator methods. The success of the proposed research approach is evident when comparing the results of the current predictive models with the existing benchmark results. Our solution could surpass the benchmark results for all the five personality traits in the ‘Big Five’ personality assessment model. It highlights the significant effect of deep semantic inspections at concept level to accurately detect individual personalities. Hence, further investigations in concept level semantic iii analysis arena is essential to move forward in the field of personality recognition than merely relying on word level analysis.