A General Personality Analysis Model Based on Social Posts and Links

Xingkong Ma, Houjie Qiu, Shujia Yao, Xinyi Chen, Jingsong Zhang, Zhaoyun Ding, Shaoyong Li, Bo Liu

Abstract: Personality plays a vital role in psychological feature analysis, product recommendation, and mental health assessment. Analyzing personality based on social networks is becoming mainstream since it allows collecting user behaviors and continuously output personality prediction results in a non-intrusive manner. However, existing methods face either over-fitting problems due to the small-sized training datasets or inaccurate feature representation due to the limited information of the testee. This paper proposes a general personality analysis model based on posts and links in social networks, called GPAM. To solve the problem of insufficient training data, we use a user linkage technique to collect large-scale and high-quality labeled personality data in a short time. By introducing posts from high-influence friends, we propose a unified personality feature extraction model to represent the users without enough information. Under various parameter settings, the experimental results demonstrate that importing moderate posts from high-influence friends benefits state-of-the-art models. The average f1-scores of predicting both MBTI and Big Five in GPAM are higher than the latest model Trignet. Compared to without introducing extra posts, the average f1-scores of in GPAM improve at least 4% for wordless users and 51% for silent users.