Personality Traits Prediction Based on Sparse Digital Footprints via Discriminative Matrix Factorization

Identifying individuals’ personality traits from their digital footprints has been proved able to improve the service of online platforms. However, due to the privacy concerns and legal restrictions, only some sparse, incomplete and anonymous digital footprints can be accessed, which seriously challenges the existing personality traits identification methods. To make the best of the available sparse digital footprints, we propose a novel personality traits prediction algorithm through jointly learning discriminative latent features for individuals and a personality traits predictor performed on the learned features. By formulating a discriminative matrix factorization problem, we seamlessly integrate the discriminative individual feature learning and personality traits predictor learning together. To solve the discriminative matrix factorization problem, we develop an alternative optimization based solution, which is efficient and easy to be parallelized for large-scale data. Experiments are conducted on the real-world Facebook like digital footprints. The results show that the proposed algorithm outperforms the state-of-the-art personality traits prediction methods significantly.