Comparison Of Machine Learning Algorithms For Personality Detection In Online Social Networking
Loading...
Date
2017-05
Authors
Sagadevan, Saravanan
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
The availability of social networks data offered opportunities to study the personality traits that advocate the behavior of criminals by applying automatic personality detection strategies. Most of the current Automatic Personality Detection (APD) studies focused on Big 5 personality model and Automatic Personality Recognization (APR) techniques to study the personalities of social network users. However, the nature of Big 5 personality model that included all the negative states under Neuroticism is not suitable for this study is due to the presence of diversities in negative states. As an alternative, this study intended to incorporate the Three Factor Personality Model (PEN) and Automatic Personality Perception approaches (APP) to classify automatically the personality of Facebook and Twitter users through their textual messages. Furthermore, throughout literature review, none of the studies found to be investigated the effects of single and combination of language models and classification levels in the perspectives of criminal personalities and social networks. Thus, this study proposed a methodology to make comparisons among Naïve Bayes (NB), Sequential Minimal Optimization (SMO), K-Nearest Neighbor (KNN) and J48 by evaluating the effects of six language models and classification levels that influenced the performances of the classifiers. Comprehensively, SMO outperformed NB, KNN and J48 in the majority of the classification processes. Moreover, the classification on the unigram attribute yielded better accuracies than other language models in the most of the classification process.
Description
Keywords
Comparison of machine learning algorithms , for personality detection in online social networking