MODELING OF USER PORTRAIT THROUGH SOCIAL MEDIA

Haiqian Gu^1,2, Jie Wang³, Ziwen Wang^1,2, Bojin Zhuang³, Fei Su^1,2

1School of Information and Communication Engineering 2Beijing Key Laboratory of Network System and Network Culture Beijing University of Posts and Telecommunications, Beijing, China

3Ping An Technology (Shenzhen) Co., Ltd. mixiu@bupt.edu.cn, wangjie388@pingan.com.cn, wangziwen@bupt.edu.cn,

zhuangbojin232@pingan.com.cn, sufei@bupt.edu.cn

ABSTRACT

·

Nowadays, massive useful data of user information and social behavior have been accumulated on the Internet, providing a possibility of profiling users personality traits online. In this paper, we propose a psychological modeling method based on computational linguistic features to profile Big Five personality traits of users on Sina Weibo (a Twitter-like microblogging service in China) and their correlations with userrsquo;s social behaviors. To the best of our knowledge, this is the first research on investigating the potential relationship between profile information, social-network behaviors and personality traits of users on Sina Weibo. Our results demonstrate an effective modeling approach to understanding demographic and psychological portraits of users on social media without customer disruption, which is useful for commercial incorporations to provide better personalized products and services.

Index Terms— User portrait, social media, Big Five personality, microblog text, user behavior

INTRODUCTION

Personality has been widely studied as it both reflects and affects peoplersquo;s behavior, which exhibits promising applications in precision marketing. Recently, data-driven psychological interpretation of usersrsquo; personality has drawn much attention. Sina Weibo, as a very popular and important social media in China, are posted about 100 million microblogs every day [1]. With accumulation of such large- scale online data, conventional questionnaire-based personality measurement becomes expensive and inefficient. To address this issue, various data-driven modeling methods have been studied for user portrait computation based on user information and social behavior. Zhao et al. [2] proved the validity of SC-LIWC (Simplified Chinese version

Thanks to Chinese National Natural Science Foundation (61532018, 61471049) for funding.

978-1-5386-1737-3/18/$31.00 copy;2018 IEEE

of Linguistic Inquiry and Word Count) in detecting psychological expressions in SNS short texts and Qiu et al.[3] studied the relationship between Chinese short texts and corresponding word categories of SC-LIWC. These two results are indeed fundamental research foundation for our work. Furthermore, Golbeck et al. [4] demonstrated that public information shared on Facebook could be used to predict usersrsquo; Big Five personality. Sumner et al. [5] demonstrated that there were some links between Dark Triad constructs and Twitter usage and employed a variety of machine learning techniques to predict these constructs in users. However, both [4] and [5] ignored usersrsquo; social behaviors, and our study make up for this limitation. On the other side, Hung et al. [6] introduced a tag-based user profiling for social media recommendation, but they did not give concrete user profiles. In short, most of the previous researches on this area utilized only one type of information on social network, such as text data or profile features. In our study, we take a comprehensive consideration on correlations between Big Five personality trait dimensions and userrsquo;s information and social behavior, and successfully predict userrsquo;s personality.

The key contributions of this paper are summarized as follows:

• An effective calculation model of Big Five personality scores based on SC-LIWC word frequencies has been proposed and demonstrated, personality profiles of Sina Weibo users have been calculated;
• 6,467 valid Weibo users have been selected to compute Big Five personality scores based on their posted short texts. And the way we train the personality calculation model has also been discussed;
• Correlation analysis between Big Five personality scores and SC-LIWC word frequency features, userrsquo;s tags, userrsquo;s demographics, userrsquo;s emoticon usage and userrsquo;s behaviors has comprehensively conducted, verifying the possibility of profiling userrsquo;s personality from microblog texts and other information they share on Sina Weibo.

THEORETICAL BACKGROUND

• 1. The Big Five personality

The Big Five model consisting of five categorical personality traits, i.e. Openness, Conscientiousness, Extraversion, Agreeableness and Neuroticism, has become one of the most widely-adopted psychological analysis models recently [7].

The Big Five model is characterized as shown in Table 1 [8]. It has well examined and developed as an important psychometric method by many researchers, which providing a comprehensive profile of an individualrsquo;s cognitive patterns.

Table 1. The dimensions of Big Five model